CISA is now turning to Anthropic’s AI to make government software safer. According to reports, the agency is using Anthropic’s AI model called Mythos to scan federal code and spot vulnerabilities before hackers can exploit them. 


CISA using Mythos to scan government software: Report

<p>Anthropic has restored global access to its advanced Claude Fable 5 and Claude Mythos 5 AI models after introducing a new set of cybersecurity safeguards. The move follows discussions with the US government and comes after the company temporarily restricted access to the models over concerns about potential misuse.</p><p>&nbsp;</p><p>The company says the updated safeguards are designed to prevent harmful cybersecurity activities while allowing legitimate users to continue using the models for coding, research and other everyday tasks. For most users, Anthropic says the experience should remain largely unchanged.</p><p>&nbsp;</p>

<blockquote class="twitter-tweet"><p lang="en" dir="ltr">Fable 5 is back. <a href="https://t.co/9RTGUCcPHy">pic.twitter.com/9RTGUCcPHy</a></p>&mdash; Claude (@claudeai) <a href="https://x.com/claudeai/status/2072402636813607381?ref_src=twsrc%5Etfw">July 1, 2026</a></blockquote> <script async src="https://platform.x.com/widgets.js" charset="utf-8"></script>

<h2><strong>Why did Anthropic bring Fable 5 back?</strong></h2><p>According to Anthropic, the company has improved its security systems to better distinguish between legitimate requests and potentially harmful ones. The new safeguards use updated classifiers that identify requests related to advanced cybersecurity misuse. If a request is flagged, users will receive a notification and the conversation will automatically switch to Claude Opus 4.8, rather than Fable 5.</p><p>&nbsp;</p><p>Anthropic says this approach allows it to continue offering powerful AI capabilities while reducing the risk of abuse.</p><p>&nbsp;</p><h2><strong>Will coding and everyday work be affected?</strong></h2><p>Anthropic says the vast majority of coding work will not be affected. However, the company acknowledged that the new safeguards may temporarily block a slightly higher number of harmless requests than intended. This means some legitimate coding or debugging prompts could be redirected to Opus 4.8 until the classifiers become more accurate.</p><p>The company said it is already working on updates to reduce these false positives over the coming weeks.</p><h2><strong>Biology and chemistry restrictions remain</strong></h2><p>Anthropic also confirmed that its biology and chemistry safety filters remain unchanged from the initial launch.</p><p>The company admitted that these classifiers are still broader than it would like. As a result, even some basic biology-related questions may currently fall back to Claude Opus 4.8 instead of Fable 5. Anthropic says improvements to these scientific safety systems are already being developed and will be introduced in future updates.</p><h2><strong>AI safety becomes a bigger industry focus</strong></h2><p>The return of Fable 5 and Mythos 5 reflects a wider shift across the AI industry, where companies are trying to balance powerful capabilities with stronger safety measures. Rather than limiting access entirely, Anthropic is introducing more targeted safeguards that aim to stop misuse while keeping advanced AI available for developers, researchers and businesses. The company says it will continue refining these protections as more users begin using the restored models.</p>

Anthropic has restored access to its Claude Fable 5 and Mythos 5 AI models after introducing updated cybersecurity safeguards. 

Anthropic has restored global access to its advanced Claude Fable 5 and Claude Mythos 5 AI models after introducing a new set of cybersecurity safeguards. The move follows discussions with the US government and comes after the company temporarily restricted access to the models over concerns about potential misuse. The company says the updated safeguards are designed to prevent harmful cybersecurity activities while allowing legitimate users to continue using the models for coding, research and other everyday tasks. For most users, Anthropic says the experience should remain largely unchanged.

<p>Anthropic has restored access to its Claude Fable 5 and Mythos 5 AI models after introducing updated cybersecurity safeguards.&nbsp;</p>

Fable 5 is available again, but with smarter safety checks: Here's what has changed for users

A demonstration of an advanced AI model has sparked fresh concerns about the risks and capabilities of next-generation artificial intelligence. During a briefing for U.S. lawmakers, an AI system reportedly showed it could identify vulnerabilities in banking and financial networks, then either help fix them or potentially exploit them.

Mythos AI demo raises cybersecurity concerns 

<p>US President Donald Trump's approach to AI regulation seems like putting a guardrail on the Silicon Valley AI giants or like requesting them to put guardrails on themselves. The framework reflects a "light touch" federal regulatory approach that emphasises innovation, preemption of state AI laws, and reliance on existing legal regimes.&nbsp;</p><p>“Advanced AI capabilities make our Nation stronger, but also introduce new national security considerations that require coordinated action across executive departments and agencies,” the order says. It was a direct contradiction to US President Donald Trump's promise of removing any guardrails before taking office. Eventhough his baseline philosophy still remains pro laissez-faire, Trump’s order “promoting advanced artificial intelligence innovation and security,” released on June 2, establishes a framework for the federal government to vet the national security risks of the most advanced AI systems, and encourages voluntary cooperation between companies and federal agencies. The keyword to be noted is "voluntary".&nbsp;</p><h2>The launch of Claude “Mythos”</h2><p>Following the launch of the Claude “Mythos” by Anthropic on April 7, the US Treasury Secretary Scott Bessent and outgoing Fed chair Jerome Powell held an emergency meeting with the Wall Street executives. Anthropic released a report titled “Assessing Claude Mythos Preview's cybersecurity capabilities.” Mythos, within a few weeks of automated and human testing, successfully identified over 10,000 high or critical-severity security flaws in major software systems, setting off alarms in Washington. The meeting was held in mid-April but was not about interest rates, not over tariffs, nor oil prices due to the Iran war, but specifically over cybersecurity risks linked to Anthropic's Mythos AI model. The attendees included the CEOs of Citigroup, Morgan Stanley, Bank of America, Wells Fargo, and Goldman Sachs. It was hastily arranged, not pre-planned. In the meeting, they were briefed about the security risks, which created a global alarm, even in India.&nbsp;</p><p><strong>Also Read: </strong><a href="https://www.wionews.com/world/xai-grok-deepfake-lawsuit-plaintiffs-identities-court-motion-1780567491628"><strong>xAI seeks identities of plaintiffs in Grok deepfake lawsuit involving sexualised images</strong></a></p><p>On April 23, Finance Minister Nirmala Sitharaman chaired a meeting with Reserve Bank of India (RBI), National Payments Corporation of India, Indian Banks’ Association, and Indian Computer Emergency Response Team officials, calling the Mythos threat “unprecedented.” Following a meeting with Anthropic's India team, officials appeared to have agreed that models used in the sensitive sector must be hosted on Indian servers for regulatory, security, and jurisdiction risks. The day after the meeting, Anthropic revealed it was releasing a version of Mythos to a select group of companies under an initiative called Project Glasswing. Participants include Amazon, Apple, and JPMorgan, with the idea that these firms will get a head start on strengthening their defences before the model becomes more widely available.&nbsp;</p><h2>Tech CEO lobby and the executive order&nbsp;</h2><p>&nbsp;</p><p>On May 21, US President Donald Trump cancelled an executive order signing, citing that “he didn't like certain aspects of it”. Just hours before the signing of the executive order, David Sacks, Trump's former crypto and AI czar, intervened against the 90-day mandatory review window. He warned that a three-month government-mandated freeze would cripple the fast-moving tech sector and hand a massive advantage to China. After the cancellation, another round of frantic negotiation started, which had all the frontier model executives and legacy tech giants, including Sam Altman, Dario Amodei, Sundar Pichai, Satya Nadella, Elon Musk, Mark Zuckerberg, Jensan Heung, Tim Cook and David Sacks, to name a prominent few. They lobbied to change the 90-day “mandatory” blocking window and reduced it to a 30-day "voluntary" review, to salvage the security framework, while keeping the competitive edge.</p><p>Anthropic welcomed Trump's executive order; Sam Altman publicly praised the compromise, stating that the executive order "gets the balance right". Arvind Krishna, CEO of IBM, said that it hit the “goldilocks spot”. The Tech CEOs intervened at the last minute to get rid of the European-style regulatory framework and established a laissez-faire framework with more autonomy. However, cybersecurity executives remained deeply wary of the practical implications of the executive order's voluntary framework, which lacks any real accountability. &nbsp;The sequence of events that led to the signing of the executive order shows the tug of war happening within the White House. It appears that the free-market crowd won this round, but the fact that an order was passed shows that a populist-regulator tension remains alive.</p><p>&nbsp;</p><p>&nbsp;</p><p>&nbsp;</p>

Donald Trump's new AI executive order promotes voluntary security reviews and a light-touch regulatory approach. The framework follows lobbying by major AI and tech companies seeking to avoid mandatory oversight.

US President Donald Trump's approach to AI regulation seems like putting a guardrail on the Silicon Valley AI giants or like requesting them to put guardrails on themselves. The framework reflects a "light touch" federal regulatory approach that emphasises innovation, preemption of state AI laws, and reliance on existing legal regimes. “Advanced AI capabilities make our Nation stronger, but also introduce new national security considerations that require coordinated action across executive departments and agencies,” the order says. It was a direct contradiction to US President Donald Trump's promise of removing any guardrails before taking office. Eventhough his baseline philosophy still remains pro laissez-faire, Trump’s order “promoting advanced artificial intelligence innovation and security,” released on June 2, establishes a framework for the federal government to vet the national security risks of the most advanced AI systems, and encourages voluntary cooperation between companies and federal agencies. The keyword to be noted is "voluntary". The launch of Claude “Mythos” Following the launch of the Claude “Mythos” by Anthropic on April 7, the US Treasury Secretary Scott Bessent and outgoing Fed chair Jerome Powell held an emergency meeting with the Wall Street executives. Anthropic released a report titled “Assessing Claude Mythos Preview's cybersecurity capabilities.” Mythos, within a few weeks of automated and human testing, successfully identified over 10,000 high or critical-severity security flaws in major software systems, setting off alarms in Washington. The meeting was held in mid-April but was not about interest rates, not over tariffs, nor oil prices due to the Iran war, but specifically over cybersecurity risks linked to Anthropic's Mythos AI model. The attendees included the CEOs of Citigroup, Morgan Stanley, Bank of America, Wells Fargo, and Goldman Sachs. It was hastily arranged, not pre-planned. In the meeting, they were briefed about the security risks, which created a global alarm, even in India. Also Read: xAI seeks identities of plaintiffs in Grok deepfake lawsuit involving sexualised images On April 23, Finance Minister Nirmala Sitharaman chaired a meeting with Reserve Bank of India (RBI), National Payments Corporation of India, Indian Banks’ Association, and Indian Computer Emergency Response Team officials, calling the Mythos threat “unprecedented.” Following a meeting with Anthropic's India team, officials appeared to have agreed that models used in the sensitive sector must be hosted on Indian servers for regulatory, security, and jurisdiction risks. The day after the meeting, Anthropic revealed it was releasing a version of Mythos to a select group of companies under an initiative called Project Glasswing. Participants include Amazon, Apple, and JPMorgan, with the idea that these firms will get a head start on strengthening their defences before the model becomes more widely available. Tech CEO lobby and the executive order On May 21, US President Donald Trump cancelled an executive order signing, citing that “he didn't like certain aspects of it”. Just hours before the signing of the executive order, David Sacks, Trump's former crypto and AI czar, intervened against the 90-day mandatory review window. He warned that a three-month government-mandated freeze would cripple the fast-moving tech sector and hand a massive advantage to China. After the cancellation, another round of frantic negotiation started, which had all the frontier model executives and legacy tech giants, including Sam Altman, Dario Amodei, Sundar Pichai, Satya Nadella, Elon Musk, Mark Zuckerberg, Jensan Heung, Tim Cook and David Sacks, to name a prominent few. They lobbied to change the 90-day “mandatory” blocking window and reduced it to a 30-day "voluntary" review, to salvage the security framework, while keeping the competitive edge. Anthropic welcomed Trump's executive order; Sam Altman publicly praised the compromise, stating that the executive order "gets the balance right". Arvind Krishna, CEO of IBM, said that it hit the “goldilocks spot”. The Tech CEOs intervened at the last minute to get rid of the European-style regulatory framework and established a laissez-faire framework with more autonomy. However, cybersecurity executives remained deeply wary of the practical implications of the executive order's voluntary framework, which lacks any real accountability. The sequence of events that led to the signing of the executive order shows the tug of war happening within the White House. It appears that the free-market crowd won this round, but the fact that an order was passed shows that a populist-regulator tension remains alive.

<p>Donald Trump's new AI executive order promotes voluntary security reviews and a light-touch regulatory approach. The framework follows lobbying by major AI and tech companies seeking to avoid mandatory oversight.</p>

Is it Wall Street vs Silicon Valley - Trump's AI regulation a battle within populists

<p>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.</p><h2><strong>What access the EU will get</strong></h2><p>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.</p><h2><strong>Why Anthropic’s Mythos is not yet available</strong></h2><p>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.</p><p>&nbsp;</p><p>This difference highlights varying approaches among AI companies when it comes to sharing advanced cybersecurity tools with regulators.</p><p>&nbsp;</p><h2><strong>Growing focus on AI in cybersecurity</strong></h2><p>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.</p><p>By granting access to the EU, OpenAI is positioning its model as a tool that can be monitored and used responsibly by trusted institutions.</p><p>&nbsp;</p><h2><strong>What OpenAI has said</strong></h2><p>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.</p><p>&nbsp;</p><h2><strong>What this means for Europe</strong></h2><p>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.</p><h2><strong>What happens next</strong></h2><p>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</p>

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

<p>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.</p>

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

<p>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.</p><h2><strong>AI finds long-hidden bugs in Firefox</strong></h2><p>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.</p><p>&nbsp;</p><p>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.</p><h2><strong>Bug fixes rise sharply with AI support</strong></h2><p>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.</p><p>&nbsp;</p><p>&nbsp;</p>

Bug fixes rise sharply with AI support

<h2><strong>AI helps find bugs, but humans still fix them</strong></h2><p>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.</p><p>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.</p><p>&nbsp;</p><h2><strong>A shift in cybersecurity approach</strong></h2><p>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.</p><p>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.</p><p>&nbsp;</p><p>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.</p><h2><strong>What this means going forward</strong></h2><p>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.</p><p>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</p><p>&nbsp;</p>

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.

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

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

<p>Anthropic has revealed details about a new experimental AI model called Claude Mythos, which the company says is its most advanced system so far. According to Anthropic, the model has demonstrated a strong ability to detect serious cybersecurity flaws in widely used software systems. Because of its capabilities, the company is not releasing the model publicly yet and is instead working with technology partners to fix the vulnerabilities it discovered.</p><p>The announcement reflects how AI tools are increasingly being used in cybersecurity research to identify weaknesses in complex software systems.</p><h2><strong>Claude Mythos identified thousands of software vulnerabilities</strong></h2><p>In a blog post, Anthropic said Claude Mythos Preview was able to identify thousands of previously unknown vulnerabilities in widely used software.</p><p>These include “zero-day vulnerabilities”, which are security flaws that software developers have not yet discovered. Zero-day vulnerabilities are particularly concerning because attackers can exploit them before developers have time to fix them.</p><h2><strong>Anthropic said the AI model identified vulnerabilities across:</strong></h2><p><strong>Web browsers</strong></p><p><strong>Operating systems</strong></p><p><strong>Core system software</strong></p><p>The company added that some of the vulnerabilities could potentially allow attackers to escalate access from a normal user account to full control of a system.</p><p><strong>Vulnerabilities discovered in OpenBSD and Linux systems</strong></p><p>Among the findings was a 27-year-old vulnerability in OpenBSD, an operating system widely used in security-focused infrastructure such as firewalls and servers. The model also identified several vulnerabilities in the Linux kernel, which powers the majority of servers worldwide and forms the backbone of many internet services.</p><p>Security researchers say the Linux kernel is one of the most widely analysed pieces of software in the world, which makes the discovery of new vulnerabilities particularly notable.</p><p>According to Anthropic, these flaws could potentially allow attackers to gain elevated access to systems if exploited.</p><h2><strong>Anthropic working with 40 technology companies</strong></h2><p>Because of the security implications, Anthropic said it is collaborating with more than 40 technology companies to address the vulnerabilities identified by Claude Mythos.</p><p>The goal is to fix the security issues before the model is released more broadly. This approach reflects a growing concern in the AI industry: extremely capable AI systems could potentially identify security weaknesses faster than humans, which creates both benefits and risks.</p><p>Anthropic said controlled testing with industry partners will help ensure that vulnerabilities are addressed responsibly.</p><h2><strong>Why the model is not yet publicly released</strong></h2><p>Although Claude Mythos reportedly outperforms earlier models such as Claude Opus 4.6 and other competing AI systems on several benchmarks, Anthropic is taking a cautious approach.</p><p>The company said the model’s cybersecurity capabilities mean it needs additional testing and safeguards before public deployment.</p><p>Anthropic CEO Dario Amodei has previously warned that highly capable AI models may present new risks if released without careful evaluation. Earlier reports also suggested that the model’s existence first became known after references appeared in a data leak related to Anthropic’s internal systems.</p><h2><strong>AI and the future of cybersecurity</strong></h2><p>Cybersecurity experts increasingly see artificial intelligence as a powerful tool for analysing large codebases and identifying weaknesses that human researchers may overlook.</p><p>Traditional vulnerability discovery often requires months of manual review by security teams. AI systems can scan code at scale and highlight suspicious patterns within minutes.</p><p>However, experts say this capability must be handled carefully. The same technology that identifies vulnerabilities could potentially be misused if released without safeguards. Anthropic’s decision to work with industry partners before releasing Claude Mythos suggests that AI companies are becoming more cautious about how powerful models are deployed. As AI research continues to advance, tools like Claude Mythos may play an important role in strengthening digital security across software systems used worldwide.</p><p>&nbsp;</p>

Anthropic has revealed an experimental AI system called Claude Mythos, which reportedly discovered thousands of previously unknown cybersecurity vulnerabilities, including zero-day flaws in systems such as OpenBSD and the Linux kernel. 

Anthropic has revealed details about a new experimental AI model called Claude Mythos, which the company says is its most advanced system so far. According to Anthropic, the model has demonstrated a strong ability to detect serious cybersecurity flaws in widely used software systems. Because of its capabilities, the company is not releasing the model publicly yet and is instead working with technology partners to fix the vulnerabilities it discovered. The announcement reflects how AI tools are increasingly being used in cybersecurity research to identify weaknesses in complex software systems. Claude Mythos identified thousands of software vulnerabilities In a blog post, Anthropic said Claude Mythos Preview was able to identify thousands of previously unknown vulnerabilities in widely used software. These include “zero-day vulnerabilities”, which are security flaws that software developers have not yet discovered. Zero-day vulnerabilities are particularly concerning because attackers can exploit them before developers have time to fix them. Anthropic said the AI model identified vulnerabilities across: Web browsers Operating systems Core system software The company added that some of the vulnerabilities could potentially allow attackers to escalate access from a normal user account to full control of a system. Vulnerabilities discovered in OpenBSD and Linux systems Among the findings was a 27-year-old vulnerability in OpenBSD, an operating system widely used in security-focused infrastructure such as firewalls and servers. The model also identified several vulnerabilities in the Linux kernel, which powers the majority of servers worldwide and forms the backbone of many internet services. Security researchers say the Linux kernel is one of the most widely analysed pieces of software in the world, which makes the discovery of new vulnerabilities particularly notable. According to Anthropic, these flaws could potentially allow attackers to gain elevated access to systems if exploited. Anthropic working with 40 technology companies Because of the security implications, Anthropic said it is collaborating with more than 40 technology companies to address the vulnerabilities identified by Claude Mythos. The goal is to fix the security issues before the model is released more broadly. This approach reflects a growing concern in the AI industry: extremely capable AI systems could potentially identify security weaknesses faster than humans, which creates both benefits and risks. Anthropic said controlled testing with industry partners will help ensure that vulnerabilities are addressed responsibly. Why the model is not yet publicly released Although Claude Mythos reportedly outperforms earlier models such as Claude Opus 4.6 and other competing AI systems on several benchmarks, Anthropic is taking a cautious approach. The company said the model’s cybersecurity capabilities mean it needs additional testing and safeguards before public deployment. Anthropic CEO Dario Amodei has previously warned that highly capable AI models may present new risks if released without careful evaluation. Earlier reports also suggested that the model’s existence first became known after references appeared in a data leak related to Anthropic’s internal systems. AI and the future of cybersecurity Cybersecurity experts increasingly see artificial intelligence as a powerful tool for analysing large codebases and identifying weaknesses that human researchers may overlook. Traditional vulnerability discovery often requires months of manual review by security teams. AI systems can scan code at scale and highlight suspicious patterns within minutes. However, experts say this capability must be handled carefully. The same technology that identifies vulnerabilities could potentially be misused if released without safeguards. Anthropic’s decision to work with industry partners before releasing Claude Mythos suggests that AI companies are becoming more cautious about how powerful models are deployed. As AI research continues to advance, tools like Claude Mythos may play an important role in strengthening digital security across software systems used worldwide.

<p>Anthropic has revealed an experimental AI system called Claude Mythos, which reportedly discovered thousands of previously unknown cybersecurity vulnerabilities, including zero-day flaws in systems such as OpenBSD and the Linux kernel.&nbsp;</p>

‘Claude Mythos in cybersecurity’: How Anthropic’s new AI model can find hidden software vulnerabilities

mythos

Mythos, News on Mythos, stories on Mythos, articles on Mythos, Mythos stories at Wion, Mythos news, Mythos latest news

mythos

CISA using Mythos to scan government software: Report

Fable 5 is available again, but with smarter safety checks

Mythos AI demo raises cybersecurity concerns

How Silicon Valley helped shape Trump's new AI security framework

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

How Anthropic’s Mythos has rewritten Firefox’s approach to cybersecurity

Claude Mythos: Anthropic’s AI model that can find critical software flaws