Anthropic, the AI company behind Claude, has warned that AI could start building itself and create its successor and that “full recursive self-improvement might increase the risks of humans losing control over AI systems.” A full recursive self-improvement refers to a stage where AI is able to improve itself without any human intervention and create a more powerful successor.&nbsp;&nbsp;The San Francisco-based firm also recommends a global pause on building the most powerful AI systems as the latest models are beginning to show signs they could evade human control.&nbsp;The company said that as it increasingly delegates AI developments to AI systems, there are signals that it leads to a future where “an AI system (is) capable of fully autonomously designing and developing its own successor.”&nbsp;Also Read: <a href="https://www.wionews.com/india-news/delhi-hotel-owner-helped-bangladeshi-nationals-get-fake-indian-documents-was-arrested-booked-last-year-1780654502514">Delhi hotel owner helped Bangladeshi nationals get fake Indian documents, was arrested, booked last year</a><h2>‘Average engineer producing eight times more code per day than in 2024’</h2>&nbsp;Anthropic shared that Claude now writes most of the code used to build the company’s AI systems. As a result, engineers are able to get much more work done, with the average engineer producing around eight times more code each day than in 2024. Claude is not just getting better at coding, but it is also improving at research tasks.&nbsp;&nbsp;Anthropic, an outlier among AI companies, said trends suggest AI systems are going to become much more capable ahead. “These trends have huge implications. AI that can build itself would be a major development in the history of technology—one that could bring enormous good for the world in science, healthcare, and beyond,” it said.&nbsp;<h2>‘All AI companies need to come together to decide on pause’&nbsp;</h2>&nbsp;It argued that all AI companies need to come together to decide to “pause” frontier AI development but warned that if only one company stopped, rivals would simply race ahead. Getting a real pause to work would mean all major AI companies in multiple countries agreeing to stop at the same time, under rules everyone could actually verify, Anthropic added.“Without a global coordination mechanism, companies and governments will have to make difficult decisions about safety while under competitive and geopolitical pressures,” it said.&nbsp;Also Read: <a href="https://www.wionews.com/world/futile-to-try-pressuring-pm-modi-putin-says-india-us-relations-do-not-affect-ties-with-russia-1780597993055">‘Futile to try pressuring PM Modi’: Putin says India-US relations do not affect ties with Russia</a>&nbsp;“We believe it would be good for the world to have the option to slow or temporarily pause frontier AI development to enable societal structures and alignment research to keep up with the advance of the technology,” it said.&nbsp;&nbsp;Anthropic compared the problem to nuclear arms control treaties but said it would be even harder to monitor since AI training is far easier to hide than a missile silo, and the temptation to quietly keep going would be enormous.&nbsp;<h2>Multiple challenges ahead for ‘pause’ proposal</h2>&nbsp;The AI giant’s research arm, Anthropic Institute, plans to study and help build systems that would be necessary to support a slowdown. It also plans to organise discussions involving policymakers, researchers, civil society groups and other AI firms to examine key questions around how to tackle the issue at hand. These questions include how to manage AI-related risks such as recursive self-improvement and how to improve mechanisms for coordination.&nbsp;Also Read: <a href="https://www.wionews.com/world/former-trump-nsa-john-bolton-to-plead-guilty-over-retaining-classified-documents-report-1780592120953">Former Trump NSA John Bolton to plead guilty over retaining classified documents: Report</a>&nbsp;The proposal would face an uphill battle in Washington and Silicon Valley, where US officials and tech executives have repeatedly argued that any slowdown in AI development risks handing China a decisive strategic edge in what many see as the defining technology race of the century.President Donald Trump also signed an executive order this week that allows the government 30 days to conduct a preliminary review of the most powerful US AI models before their release.&nbsp;The company stressed that such a stage has not yet arrived but argued that governments, regulators and society need to start preparing for it now.&nbsp;

Anthropic revealed that Claude now writes most of the code used to build the company’s AI systems, and as a result, engineers are able to get much more work done, with the average engineer producing around eight times more code each day than in 2024.

Anthropic, the AI company behind Claude, has warned that AI could start building itself and create its successor and that “full recursive self-improvement might increase the risks of humans losing control over AI systems.” A full recursive self-improvement refers to a stage where AI is able to improve itself without any human intervention and create a more powerful successor. The San Francisco-based firm also recommends a global pause on building the most powerful AI systems as the latest models are beginning to show signs they could evade human control. The company said that as it increasingly delegates AI developments to AI systems, there are signals that it leads to a future where “an AI system (is) capable of fully autonomously designing and developing its own successor.” Also Read: Delhi hotel owner helped Bangladeshi nationals get fake Indian documents, was arrested, booked last year ‘Average engineer producing eight times more code per day than in 2024’ Anthropic shared that Claude now writes most of the code used to build the company’s AI systems. As a result, engineers are able to get much more work done, with the average engineer producing around eight times more code each day than in 2024. Claude is not just getting better at coding, but it is also improving at research tasks. Anthropic, an outlier among AI companies, said trends suggest AI systems are going to become much more capable ahead. “These trends have huge implications. AI that can build itself would be a major development in the history of technology—one that could bring enormous good for the world in science, healthcare, and beyond,” it said. ‘All AI companies need to come together to decide on pause’ It argued that all AI companies need to come together to decide to “pause” frontier AI development but warned that if only one company stopped, rivals would simply race ahead. Getting a real pause to work would mean all major AI companies in multiple countries agreeing to stop at the same time, under rules everyone could actually verify, Anthropic added. “Without a global coordination mechanism, companies and governments will have to make difficult decisions about safety while under competitive and geopolitical pressures,” it said. Also Read: ‘Futile to try pressuring PM Modi’: Putin says India-US relations do not affect ties with Russia “We believe it would be good for the world to have the option to slow or temporarily pause frontier AI development to enable societal structures and alignment research to keep up with the advance of the technology,” it said. Anthropic compared the problem to nuclear arms control treaties but said it would be even harder to monitor since AI training is far easier to hide than a missile silo, and the temptation to quietly keep going would be enormous. Multiple challenges ahead for ‘pause’ proposal The AI giant’s research arm, Anthropic Institute, plans to study and help build systems that would be necessary to support a slowdown. It also plans to organise discussions involving policymakers, researchers, civil society groups and other AI firms to examine key questions around how to tackle the issue at hand. These questions include how to manage AI-related risks such as recursive self-improvement and how to improve mechanisms for coordination. Also Read: Former Trump NSA John Bolton to plead guilty over retaining classified documents: Report The proposal would face an uphill battle in Washington and Silicon Valley, where US officials and tech executives have repeatedly argued that any slowdown in AI development risks handing China a decisive strategic edge in what many see as the defining technology race of the century. President Donald Trump also signed an executive order this week that allows the government 30 days to conduct a preliminary review of the most powerful US AI models before their release. The company stressed that such a stage has not yet arrived but argued that governments, regulators and society need to start preparing for it now.

Anthropic revealed that Claude now writes most of the code used to build the company’s AI systems, and as a result, engineers are able to get much more work done, with the average engineer producing around eight times more code each day than in 2024.

‘Time for pause’: Humans may lose control on AI as it starts building itself, warns Anthropic

Artificial intelligence company Anthropic has announced a major expansion of Project Glasswing, its cyber security initiative designed to help organisations identify and fix critical software vulnerabilities using AI.The programme will now include around 150 organisations across more than 15 countries, significantly expanding from its initial group of participants. The move comes just a day after Anthropic confirmed that it had confidentially filed for an initial public offering (IPO), highlighting the company's growing influence in both artificial intelligence and cyber security. At the centre of Project Glasswing is Anthropic's advanced AI model, Claude Mythos, which the company says can identify thousands of serious software vulnerabilities in a matter of weeks.<h2>What is Project Glasswing?</h2>Project Glasswing was launched earlier this year as a joint industry initiative aimed at strengthening cyber security using artificial intelligence. In April, Anthropic provided access to Claude Mythos Preview to an initial group of around 50 partners, including organisations connected to the US government. The goal was to scan large software codebases and identify hidden security weaknesses before cyber criminals could exploit them.According to Anthropic, the AI model has already helped discover thousands of high-severity vulnerabilities, including flaws affecting critical digital infrastructure. The company says Project Glasswing is not just about finding bugs. It is also focused on building safety frameworks and safeguards before highly capable cyber security AI tools become widely available.<h2>Why is Anthropic expanding the programme?</h2>Anthropic says the expanded group will include organisations operating in sectors that were underrepresented during the first phase of the programme.These sectors include:Energy and power systems Water infrastructure Healthcare services Communications networks Hardware and semiconductor companiesAccording to the company, many of these organisations maintain software and digital infrastructure that millions of people rely on every day. Anthropic stated that a successful cyber attack against many of these systems could have serious consequences for national security, public services and critical infrastructure. The company estimates that a major attack on some participating organisations could potentially affect more than 100 million people.<h2>Which countries are involved?</h2>According to reports from the Financial Times, the expanded programme includes organisations from several countries that work closely with the United States.These reportedly include:Australia Canada France Germany Italy Netherlands Spain Belgium Sweden Switzerland India Japan New Zealand South KoreaThe wider international participation reflects growing concern among governments and businesses about increasingly sophisticated cyber threats. Several major organisations are also reported to be participating, including identity management company Okta, technology firms Samsung, SK Hynix and SK Telecom, along with NATO and the European Union Agency for Cybersecurity (ENISA).<h2>What makes Claude Mythos different?</h2>Claude Mythos is Anthropic's most advanced cyber security-focused AI model. The company claims the model can analyse software at a scale far beyond what traditional security teams can achieve alone. During testing, Mythos reportedly identified thousands of previously unknown vulnerabilities, including flaws affecting major operating systems and software platforms.Anthropic says the model's ability to detect zero-day vulnerabilities has made it one of the most powerful AI systems currently being tested for cyber defence applications. However, these same capabilities have also raised concerns among security experts and policymakers. Some worry that similar technology could eventually be misused if deployed without proper safeguards.That concern is one reason why Anthropic has restricted access to Mythos and is using Project Glasswing as a controlled testing environment.<h2>AI companies are entering a cyber security race</h2>Anthropic's expansion comes as competition among AI companies increasingly moves beyond chatbots and into cyber security. The company has warned that rival firms are likely to develop similarly capable systems in the near future. As a result, Anthropic is trying to establish safety measures before advanced cyber security AI becomes widely available.The competitive pressure is already visible. OpenAI recently introduced GPT-5.5-Cyber, a specialised model focused on cyber security applications, and has begun testing it with selected partners. Industry analysts believe cyber security could become one of the most important areas for advanced AI deployment over the next few years.

Anthropic has announced a major expansion of Project Glasswing, its cybersecurity initiative designed to help organisations identify and fix critical software vulnerabilities using AI.

Artificial intelligence company Anthropic has announced a major expansion of Project Glasswing, its cyber security initiative designed to help organisations identify and fix critical software vulnerabilities using AI. The programme will now include around 150 organisations across more than 15 countries, significantly expanding from its initial group of participants. The move comes just a day after Anthropic confirmed that it had confidentially filed for an initial public offering (IPO), highlighting the company's growing influence in both artificial intelligence and cyber security. At the centre of Project Glasswing is Anthropic's advanced AI model, Claude Mythos, which the company says can identify thousands of serious software vulnerabilities in a matter of weeks. What is Project Glasswing? Project Glasswing was launched earlier this year as a joint industry initiative aimed at strengthening cyber security using artificial intelligence. In April, Anthropic provided access to Claude Mythos Preview to an initial group of around 50 partners, including organisations connected to the US government. The goal was to scan large software codebases and identify hidden security weaknesses before cyber criminals could exploit them. According to Anthropic, the AI model has already helped discover thousands of high-severity vulnerabilities, including flaws affecting critical digital infrastructure. The company says Project Glasswing is not just about finding bugs. It is also focused on building safety frameworks and safeguards before highly capable cyber security AI tools become widely available. Why is Anthropic expanding the programme? Anthropic says the expanded group will include organisations operating in sectors that were underrepresented during the first phase of the programme. These sectors include: Energy and power systems Water infrastructure Healthcare services Communications networks Hardware and semiconductor companies According to the company, many of these organisations maintain software and digital infrastructure that millions of people rely on every day. Anthropic stated that a successful cyber attack against many of these systems could have serious consequences for national security, public services and critical infrastructure. The company estimates that a major attack on some participating organisations could potentially affect more than 100 million people. Which countries are involved? According to reports from the Financial Times, the expanded programme includes organisations from several countries that work closely with the United States. These reportedly include: Australia Canada France Germany Italy Netherlands Spain Belgium Sweden Switzerland India Japan New Zealand South Korea The wider international participation reflects growing concern among governments and businesses about increasingly sophisticated cyber threats. Several major organisations are also reported to be participating, including identity management company Okta, technology firms Samsung, SK Hynix and SK Telecom, along with NATO and the European Union Agency for Cybersecurity (ENISA). What makes Claude Mythos different? Claude Mythos is Anthropic's most advanced cyber security-focused AI model. The company claims the model can analyse software at a scale far beyond what traditional security teams can achieve alone. During testing, Mythos reportedly identified thousands of previously unknown vulnerabilities, including flaws affecting major operating systems and software platforms. Anthropic says the model's ability to detect zero-day vulnerabilities has made it one of the most powerful AI systems currently being tested for cyber defence applications. However, these same capabilities have also raised concerns among security experts and policymakers. Some worry that similar technology could eventually be misused if deployed without proper safeguards. That concern is one reason why Anthropic has restricted access to Mythos and is using Project Glasswing as a controlled testing environment. AI companies are entering a cyber security race Anthropic's expansion comes as competition among AI companies increasingly moves beyond chatbots and into cyber security. The company has warned that rival firms are likely to develop similarly capable systems in the near future. As a result, Anthropic is trying to establish safety measures before advanced cyber security AI becomes widely available. The competitive pressure is already visible. OpenAI recently introduced GPT-5.5-Cyber, a specialised model focused on cyber security applications, and has begun testing it with selected partners. Industry analysts believe cyber security could become one of the most important areas for advanced AI deployment over the next few years.

Anthropic has announced a major expansion of Project Glasswing, its cybersecurity initiative designed to help organisations identify and fix critical software vulnerabilities using AI.

Anthropic expands Project Glasswing to 150 organisations across 15 countries

Anthropic has revealed that its advanced AI system Claude Mythos identified more than 10,000 critical software vulnerabilities in just one month, highlighting how artificial intelligence is transforming cybersecurity. The model, which is not publicly available, is being tested under a restricted programme called Project Glasswing, where selected companies are given access to evaluate its capabilities.&nbsp;<h2>How Claude Mythos is being used</h2>Claude Mythos has been shared with a limited group of companies, including major technology firms such as Amazon and Google. Under this programme, organisations are using the AI to scan their systems for security weaknesses. According to Anthropic, many participants reported a significant increase in the number of vulnerabilities detected. In some cases, companies said their bug detection rate improved by more than ten times compared to earlier tools.<h2>Real-world results from companies</h2>Several organisations have shared early results from testing the AI system:Cloudflare found around 2,000 bugs, including 400 high or critical vulnerabilitiesMozilla identified and fixed 271 vulnerabilities in Firefox during testingThe UK AI Security Institute reported that Mythos successfully completed complex cyberattack simulations These findings suggest that AI systems like Mythos can detect issues at a scale and speed beyond traditional methods.<h2>Why the model is not publicly released</h2>Despite its capabilities, Anthropic has decided not to release Mythos to the general public. The company said that such powerful systems could be misused if they fall into the wrong hands. Governments in different countries have also raised concerns about the risks linked to advanced AI tools. Anthropic has stated that current safeguards are not strong enough to fully prevent misuse, which is why access remains restricted.<h2>Plans to expand access with governments</h2>Anthropic has said it will work with the United States and allied governments to expand access to the model in a controlled way. However, it has not confirmed which countries will be included in this next phase. The company indicated that wider access may be considered in the future if similar capabilities become common across the industry.Impact on cybersecurity and software updatesThe use of AI in bug detection is also speeding up how quickly companies fix vulnerabilities.Reports indicate that:Microsoft expects patch releases to increaseOracle is fixing issues faster across its systemsSecurity firms are releasing more updates than usualAnthropic also said it has used Mythos to scan over 1,000 open-source projects, which form a key part of global internet infrastructure.<h2>What this means for the AI industry</h2>The development shows how AI is becoming a key tool in cybersecurity. While it can improve defence systems, it also raises questions about how such technology should be controlled. The company is reportedly planning a new funding round that could raise up to $30 billion, potentially valuing Anthropic at around $900 billion, according to reports.&nbsp;What happens nextAnthropic is expected to continue testing Mythos with selected partners while exploring ways to expand access safely.The results so far show that AI can play a major role in identifying software risks. At the same time, the debate around safety and control is likely to grow as these systems become more advanced.

Anthropic has revealed that its advanced AI system Claude Mythos identified more than 10,000 critical software vulnerabilities in just one month, highlighting how artificial intelligence is transforming cybersecurity. 


Anthropic has revealed that its advanced AI system Claude Mythos identified more than 10,000 critical software vulnerabilities in just one month, highlighting how artificial intelligence is transforming cybersecurity. The model, which is not publicly available, is being tested under a restricted programme called Project Glasswing, where selected companies are given access to evaluate its capabilities. How Claude Mythos is being used Claude Mythos has been shared with a limited group of companies, including major technology firms such as Amazon and Google. Under this programme, organisations are using the AI to scan their systems for security weaknesses. According to Anthropic, many participants reported a significant increase in the number of vulnerabilities detected. In some cases, companies said their bug detection rate improved by more than ten times compared to earlier tools. Real-world results from companies Several organisations have shared early results from testing the AI system: Cloudflare found around 2,000 bugs, including 400 high or critical vulnerabilities Mozilla identified and fixed 271 vulnerabilities in Firefox during testing The UK AI Security Institute reported that Mythos successfully completed complex cyberattack simulations These findings suggest that AI systems like Mythos can detect issues at a scale and speed beyond traditional methods. Why the model is not publicly released Despite its capabilities, Anthropic has decided not to release Mythos to the general public. The company said that such powerful systems could be misused if they fall into the wrong hands. Governments in different countries have also raised concerns about the risks linked to advanced AI tools. Anthropic has stated that current safeguards are not strong enough to fully prevent misuse, which is why access remains restricted. Plans to expand access with governments Anthropic has said it will work with the United States and allied governments to expand access to the model in a controlled way. However, it has not confirmed which countries will be included in this next phase. The company indicated that wider access may be considered in the future if similar capabilities become common across the industry. Impact on cybersecurity and software updates The use of AI in bug detection is also speeding up how quickly companies fix vulnerabilities. Reports indicate that: Microsoft expects patch releases to increase Oracle is fixing issues faster across its systems Security firms are releasing more updates than usual Anthropic also said it has used Mythos to scan over 1,000 open-source projects, which form a key part of global internet infrastructure. What this means for the AI industry The development shows how AI is becoming a key tool in cybersecurity. While it can improve defence systems, it also raises questions about how such technology should be controlled. The company is reportedly planning a new funding round that could raise up to $30 billion, potentially valuing Anthropic at around $900 billion, according to reports. What happens next Anthropic is expected to continue testing Mythos with selected partners while exploring ways to expand access safely. The results so far show that AI can play a major role in identifying software risks. At the same time, the debate around safety and control is likely to grow as these systems become more advanced.

Anthropic has revealed that its advanced AI system Claude Mythos identified more than 10,000 critical software vulnerabilities in just one month, highlighting how artificial intelligence is transforming cybersecurity.&nbsp;

‘10,000 bugs in a month’: How Anthropic’s AI is changing cybersecurity

OpenAI has said it will provide the European Union with access to its new cybersecurity-focused AI model, GPT-5.5-Cyber, while Anthropic has not yet extended similar access for its Mythos model. The announcement marks a key development in how advanced AI tools are being shared with governments and institutions. According to OpenAI, the model will be available in a limited preview to selected partners, including businesses, cyber agencies and EU bodies.<h2>What access the EU will get</h2>European organisations, including the EU AI Office and cybersecurity authorities, will be able to test and review GPT-5.5-Cyber. The aim is to allow regulators and experts to closely monitor how the model works and how it could be used in real-world security scenarios. Officials said this access would help address concerns around safety and ensure that the deployment of such tools is properly understood. According to EU spokesperson Thomas Regnier, discussions between OpenAI and the European Commission are ongoing, with further meetings planned.<h2>Why Anthropic’s Mythos is not yet available</h2>While OpenAI has moved ahead with access, Anthropic has not yet provided the EU with preview access to its Mythos model. The EU has confirmed that discussions with Anthropic are taking place, but they are still at an earlier stage. Officials said that while several meetings have been held, no agreement similar to OpenAI’s current arrangement has been reached so far.&nbsp;This difference highlights varying approaches among AI companies when it comes to sharing advanced cybersecurity tools with regulators.&nbsp;<h2>Growing focus on AI in cybersecurity</h2>The move comes at a time when AI tools are increasingly being used in cybersecurity. Models like GPT-5.5-Cyber are designed to help identify vulnerabilities, analyse threats and support defence systems. Recent developments, including the release of Mythos, have raised concerns about how powerful such tools could be, especially if used without proper oversight.By granting access to the EU, OpenAI is positioning its model as a tool that can be monitored and used responsibly by trusted institutions.&nbsp;<h2>What OpenAI has said</h2>OpenAI officials have said that cybersecurity should not be managed by companies alone. The company’s Head of OpenAI for Countries, George Osborne, stated that collaboration with governments and institutions is important to ensure safety. He added that making such tools available to trusted partners can help improve security and support public safety efforts.&nbsp;<h2>What this means for Europe</h2>Access to GPT-5.5-Cyber could allow European institutions to better understand how AI can be used in defence systems. It may also help shape future policies around AI and cybersecurity. At the same time, the delay in accessing Anthropic’s Mythos shows that not all AI tools are being shared at the same pace.<h2>What happens next</h2>Further discussions between the EU and AI companies are expected in the coming weeks. The outcome of these talks could influence how advanced AI systems are regulated and deployed across Europe. As cybersecurity risks continue to grow, access to such tools may become an important part of how governments prepare for future threats

OpenAI has said it will provide the European Union with access to its new cybersecurity-focused AI model, GPT-5.5-Cyber, while Anthropic has not yet extended similar access for its Mythos model.


OpenAI has said it will provide the European Union with access to its new cybersecurity-focused AI model, GPT-5.5-Cyber, while Anthropic has not yet extended similar access for its Mythos model. The announcement marks a key development in how advanced AI tools are being shared with governments and institutions. According to OpenAI, the model will be available in a limited preview to selected partners, including businesses, cyber agencies and EU bodies. What access the EU will get European organisations, including the EU AI Office and cybersecurity authorities, will be able to test and review GPT-5.5-Cyber. The aim is to allow regulators and experts to closely monitor how the model works and how it could be used in real-world security scenarios. Officials said this access would help address concerns around safety and ensure that the deployment of such tools is properly understood. According to EU spokesperson Thomas Regnier, discussions between OpenAI and the European Commission are ongoing, with further meetings planned. Why Anthropic’s Mythos is not yet available While OpenAI has moved ahead with access, Anthropic has not yet provided the EU with preview access to its Mythos model. The EU has confirmed that discussions with Anthropic are taking place, but they are still at an earlier stage. Officials said that while several meetings have been held, no agreement similar to OpenAI’s current arrangement has been reached so far. This difference highlights varying approaches among AI companies when it comes to sharing advanced cybersecurity tools with regulators. Growing focus on AI in cybersecurity The move comes at a time when AI tools are increasingly being used in cybersecurity. Models like GPT-5.5-Cyber are designed to help identify vulnerabilities, analyse threats and support defence systems. Recent developments, including the release of Mythos, have raised concerns about how powerful such tools could be, especially if used without proper oversight. By granting access to the EU, OpenAI is positioning its model as a tool that can be monitored and used responsibly by trusted institutions. What OpenAI has said OpenAI officials have said that cybersecurity should not be managed by companies alone. The company’s Head of OpenAI for Countries, George Osborne, stated that collaboration with governments and institutions is important to ensure safety. He added that making such tools available to trusted partners can help improve security and support public safety efforts. What this means for Europe Access to GPT-5.5-Cyber could allow European institutions to better understand how AI can be used in defence systems. It may also help shape future policies around AI and cybersecurity. At the same time, the delay in accessing Anthropic’s Mythos shows that not all AI tools are being shared at the same pace. What happens next Further discussions between the EU and AI companies are expected in the coming weeks. The outcome of these talks could influence how advanced AI systems are regulated and deployed across Europe. As cybersecurity risks continue to grow, access to such tools may become an important part of how governments prepare for future threats

OpenAI has said it will provide the European Union with access to its new cybersecurity-focused AI model, GPT-5.5-Cyber, while Anthropic has not yet extended similar access for its Mythos model.

OpenAI pushes cyber AI into Europe but Anthropic still not releasing Mythos

Anthropic’s new Mythos model is changing how companies approach cybersecurity, with early results showing a sharp improvement in detecting software vulnerabilities. When the model was introduced in April 2026, it reportedly identified thousands of high-severity bugs, many of which needed to be fixed before the system could be made widely available. Now, researchers working on Mozilla Firefox have shared how this technology is already being used in practice.<h2>AI finds long-hidden bugs in Firefox</h2>According to Mozilla’s security team, Mythos has uncovered several serious vulnerabilities, including issues that had remained undetected for more than a decade. This marks a significant shift compared to earlier AI tools, which often produced large numbers of low-quality alerts and false positives.&nbsp;Researchers say that recent improvements in AI systems, especially those that can review and refine their own outputs, have made these tools more reliable. As a result, security teams are now able to focus on meaningful findings instead of filtering out incorrect reports.<h2>Bug fixes rise sharply with AI support</h2>The impact of AI tools is clearly visible in Firefox’s recent updates. In April 2026, the browser released 423 bug fixes, compared to just 31 fixes in the same month last year. This increase shows how AI is helping teams identify more issues in less time. Some of these bugs were highly complex, including vulnerabilities in Firefox’s sandbox system. These types of issues are difficult to detect because they require multiple steps, including creating test code and simulating attacks on secure parts of the browser.&nbsp;&nbsp;

Bug fixes rise sharply with AI support

<h2>AI helps find bugs, but humans still fix them</h2>Despite these advances, human engineers remain central to the process. While AI can suggest possible fixes, the final code is still written and reviewed by developers.Mozilla’s team has said that AI-generated patches often need refinement and cannot be directly applied. For now, the role of AI is mainly to assist in discovery rather than fully automate the repair process.&nbsp;<h2>A shift in cybersecurity approach</h2>The use of tools like Mythos reflects a broader change in cybersecurity. Instead of relying only on manual testing or traditional scanning tools, companies are now using AI to analyse code more deeply and identify hidden risks.However, experts also note that these tools could be used by both defenders and attackers. While AI can help fix vulnerabilities faster, it may also enable malicious actors to discover weaknesses more quickly. Dario Amodei has suggested that, if managed properly, these tools could improve overall security by helping organisations fix vulnerabilities before they are exploited.&nbsp;At the same time, Mozilla engineers have taken a cautious view, noting that the long-term impact is still unclear. They believe that while AI gives an advantage to defenders, the balance between attackers and defenders may continue to evolve.<h2>What this means going forward</h2>The use of AI in cybersecurity is growing rapidly, and tools like Mythos show how quickly capabilities are improving. The ability to detect complex and long-standing bugs could help make software more secure.At the same time, the technology raises new questions about how security risks will be managed in the future. As AI tools become more widely available, companies will need to adapt their strategies to keep systems safe while making the most of these new capabilities&nbsp;

Anthropic’s new Mythos model is changing how companies approach cybersecurity, with early results showing a sharp improvement in detecting software vulnerabilities. 

Anthropic’s new Mythos model is changing how companies approach cybersecurity, with early results showing a sharp improvement in detecting software vulnerabilities. When the model was introduced in April 2026, it reportedly identified thousands of high-severity bugs, many of which needed to be fixed before the system could be made widely available. Now, researchers working on Mozilla Firefox have shared how this technology is already being used in practice. AI finds long-hidden bugs in Firefox According to Mozilla’s security team, Mythos has uncovered several serious vulnerabilities, including issues that had remained undetected for more than a decade. This marks a significant shift compared to earlier AI tools, which often produced large numbers of low-quality alerts and false positives. Researchers say that recent improvements in AI systems, especially those that can review and refine their own outputs, have made these tools more reliable. As a result, security teams are now able to focus on meaningful findings instead of filtering out incorrect reports. Bug fixes rise sharply with AI support The impact of AI tools is clearly visible in Firefox’s recent updates. In April 2026, the browser released 423 bug fixes, compared to just 31 fixes in the same month last year. This increase shows how AI is helping teams identify more issues in less time. Some of these bugs were highly complex, including vulnerabilities in Firefox’s sandbox system. These types of issues are difficult to detect because they require multiple steps, including creating test code and simulating attacks on secure parts of the browser.

Anthropic’s new Mythos model is changing how companies approach cybersecurity, with early results showing a sharp improvement in detecting software vulnerabilities.&nbsp;

‘Anthropic’s Mythos AI bug detection’: What it means for the future of software security

Artificial intelligence is increasingly being used to strengthen financial security systems. Several major banks in the United States are now experimenting with a new AI model developed by Anthropic to identify potential vulnerabilities in their digital infrastructure.According to reports, senior US officials recently encouraged financial institutions to test Anthropic’s new Mythos model to see whether it can detect weaknesses in complex banking systems.The discussions reportedly involved top officials including Jerome Powell and Scott Bessent, who met leading bank executives to explore how artificial intelligence could strengthen cybersecurity and risk detection.<h2>Major US banks quietly testing the AI model</h2>According to Bloomberg, several large financial institutions are already experimenting with the Mythos model. Among the banks involved is JPMorgan Chase, which has been identified as one of the early partners testing the technology. Other institutions reportedly examining the model include Goldman Sachs, Citigroup, Bank of America, and Morgan Stanley. The goal of these trials is to evaluate whether AI can help detect security vulnerabilities before they develop into real threats. Interestingly, the Mythos system was not specifically designed as a cybersecurity tool. However, early testing suggests it may be effective at analysing complex systems and identifying hidden weaknesses.<h2>What makes the Mythos model different?</h2>The Mythos AI model was developed by Anthropic, a company based in San Francisco and widely seen as a rival to OpenAI. Instead of focusing only on chat or content generation, Mythos appears capable of analysing large technical systems and identifying potential vulnerabilities.Experts say this type of AI capability could be useful in sectors where digital infrastructure is complex and sensitive, such as banking and financial services. However, some industry observers have cautioned that the technology is still new and may not yet be ready for large-scale deployment.<h2>Controversy surrounding Anthropic</h2>Anthropic has also faced scrutiny in recent months. According to reports, the company was labelled a supply-chain risk by the US Department of Defense following disagreements over how its AI models could be used.Concerns have also emerged outside the United States. Reports from the Financial Times suggest regulators in the United Kingdom are reviewing potential risks related to the Mythos system. These discussions highlight the growing debate about how advanced AI models should be governed and monitored.<h2>Claude Code leak raises questions</h2>In a separate development, Anthropic recently faced an unexpected security issue involving its coding tool, Claude Code. A debugging file was mistakenly included in a public package, exposing internal details about the system. According to reports cited by VentureBeat, roughly 512,000 lines of TypeScript code were visible to developers who accessed the files.The leak reportedly revealed internal features designed to improve how the system handles long conversations, including a mechanism called “Self-Healing Memory”, which helps the AI manage context more effectively. Although the issue was quickly noticed by developers online, it raised questions about how AI companies manage internal development tools.<h2>Growing role of AI in financial security</h2>The testing of Mythos highlights a wider trend in the technology sector. Banks and financial institutions are increasingly exploring AI systems to improve risk detection, fraud prevention and cybersecurity.The financial industry operates some of the world’s most complex digital networks, making early detection of vulnerabilities a priority. While it remains unclear how widely the Mythos model will eventually be adopted, the interest from major banks shows how quickly AI is becoming part of financial infrastructure. For regulators and technology companies alike, the challenge now is to balance innovation with security and oversight as AI tools continue to evolve.

Major US banks are testing Claude Mythos, developed by Anthropic, to detect cybersecurity weaknesses in banking infrastructure. Institutions including JPMorgan Chase, Goldman Sachs, Citigroup, Bank of America, and Morgan Stanley are reportedly evaluating the system. 

Artificial intelligence is increasingly being used to strengthen financial security systems. Several major banks in the United States are now experimenting with a new AI model developed by Anthropic to identify potential vulnerabilities in their digital infrastructure. According to reports, senior US officials recently encouraged financial institutions to test Anthropic’s new Mythos model to see whether it can detect weaknesses in complex banking systems. The discussions reportedly involved top officials including Jerome Powell and Scott Bessent, who met leading bank executives to explore how artificial intelligence could strengthen cybersecurity and risk detection. Major US banks quietly testing the AI model According to Bloomberg, several large financial institutions are already experimenting with the Mythos model. Among the banks involved is JPMorgan Chase, which has been identified as one of the early partners testing the technology. Other institutions reportedly examining the model include Goldman Sachs, Citigroup, Bank of America, and Morgan Stanley. The goal of these trials is to evaluate whether AI can help detect security vulnerabilities before they develop into real threats. Interestingly, the Mythos system was not specifically designed as a cybersecurity tool. However, early testing suggests it may be effective at analysing complex systems and identifying hidden weaknesses. What makes the Mythos model different? The Mythos AI model was developed by Anthropic, a company based in San Francisco and widely seen as a rival to OpenAI. Instead of focusing only on chat or content generation, Mythos appears capable of analysing large technical systems and identifying potential vulnerabilities. Experts say this type of AI capability could be useful in sectors where digital infrastructure is complex and sensitive, such as banking and financial services. However, some industry observers have cautioned that the technology is still new and may not yet be ready for large-scale deployment. Controversy surrounding Anthropic Anthropic has also faced scrutiny in recent months. According to reports, the company was labelled a supply-chain risk by the US Department of Defense following disagreements over how its AI models could be used. Concerns have also emerged outside the United States. Reports from the Financial Times suggest regulators in the United Kingdom are reviewing potential risks related to the Mythos system. These discussions highlight the growing debate about how advanced AI models should be governed and monitored. Claude Code leak raises questions In a separate development, Anthropic recently faced an unexpected security issue involving its coding tool, Claude Code. A debugging file was mistakenly included in a public package, exposing internal details about the system. According to reports cited by VentureBeat, roughly 512,000 lines of TypeScript code were visible to developers who accessed the files. The leak reportedly revealed internal features designed to improve how the system handles long conversations, including a mechanism called “Self-Healing Memory”, which helps the AI manage context more effectively. Although the issue was quickly noticed by developers online, it raised questions about how AI companies manage internal development tools. Growing role of AI in financial security The testing of Mythos highlights a wider trend in the technology sector. Banks and financial institutions are increasingly exploring AI systems to improve risk detection, fraud prevention and cybersecurity. The financial industry operates some of the world’s most complex digital networks, making early detection of vulnerabilities a priority. While it remains unclear how widely the Mythos model will eventually be adopted, the interest from major banks shows how quickly AI is becoming part of financial infrastructure. For regulators and technology companies alike, the challenge now is to balance innovation with security and oversight as AI tools continue to evolve.

Major US banks are testing Claude Mythos, developed by Anthropic, to detect cybersecurity weaknesses in banking infrastructure. Institutions including JPMorgan Chase, Goldman Sachs, Citigroup, Bank of America, and Morgan Stanley are reportedly evaluating the system.&nbsp;

Claude Mythos AI is being tested by major US banks for system vulnerabilities, reports say

Anthropic’s Claude Mythos Preview showed advanced ability to find and exploit vulnerabilities across computer systems. This is raising fears of AI-driven cyberattacks and financial instability. The US government has issued warnings to banking and financial industry leaders. Here is why it matters

Anthropic’s Claude Mythos Preview showed advanced ability to find and exploit bugs across computer systems. This is raising fears of AI-driven cyberattacks and financial instability. The US government has issued warnings to banking and financial industry leaders. Here is why it matters

Why Anthropic’s Claude Mythos is scaring financial sector and US officials: This AI model can end cybersecurity, if in wrong hands