Written by Silvia Pavelli
Anthropic restricts access to its latest AI model, “Mythos,” capable of autonomously exploiting vulnerabilities across major operating systems and web browsers, triggering urgent warnings from US officials and a massive selloff in SaaS stocks.
In an unprecedented move that has sent shockwaves through the cybersecurity and financial sectors, AI research company Anthropic has announced the indefinite restriction of public access to its latest frontier model, code-named “Claude Mythos.” The decision comes after internal red-teaming revealed the model possesses an alarming capacity to autonomously discover and exploit zero-day vulnerabilities in every major operating system and web browser currently in use.
The disclosure of Mythos’s capabilities has triggered an immediate and severe reaction from the US government and global financial markets, highlighting the escalating stakes in the race for artificial general intelligence (AGI) and the profound security implications of advanced AI systems.
The Capabilities of Claude Mythos
According to sources familiar with Anthropic’s internal testing, Claude Mythos represents a significant leap forward in autonomous reasoning and code execution. Unlike previous models that required detailed human prompting to identify potential security flaws, Mythos demonstrated the ability to independently map network topologies, identify unpatched vulnerabilities, and write custom exploit code without human intervention.
During a simulated cyber-attack scenario, the model successfully breached heavily fortified test environments running the latest versions of Windows, macOS, Linux, Chrome, and Safari. The speed and efficiency with which Mythos executed these attacks reportedly alarmed even Anthropic’s most seasoned safety researchers.
“The model’s ability to chain together complex, multi-stage exploits autonomously is unlike anything we have seen before,” stated a senior cybersecurity analyst who reviewed the red-teaming report. “It essentially automates the role of an advanced persistent threat (APT) group, operating at machine speed.”
Urgent Warnings and Market Turmoil
The potential for such a tool to fall into the hands of malicious actors prompted immediate action from the highest levels of the US government. Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell convened an emergency briefing with the CEOs of major US banks, issuing stark warnings about the systemic risks posed by AI-driven cyber threats.
The briefing emphasized the need for financial institutions to immediately harden their digital infrastructure and review their incident response protocols. The unprecedented nature of the warning from top financial regulators underscored the perceived severity of the threat.
The news of Mythos’s capabilities and the subsequent government warnings had an immediate and devastating impact on the stock market, particularly within the Software-as-a-Service (SaaS) sector. Investors, fearing that widespread vulnerabilities could compromise cloud-based platforms and enterprise software, initiated a massive selloff.
Major SaaS providers saw their stock prices plummet by double digits within hours of the news breaking. The broader tech-heavy NASDAQ index also experienced a significant decline, reflecting widespread anxiety about the security of the digital economy in the age of advanced AI.
Anthropic’s Response and the Path Forward
In a statement released late Wednesday, Anthropic CEO Dario Amodei defended the decision to restrict access to Mythos, emphasizing the company’s commitment to responsible AI development.
“The capabilities demonstrated by Claude Mythos during our safety evaluations clearly cross the threshold of acceptable risk for broad deployment,” Amodei stated. “We believe that releasing this model in its current form would be deeply irresponsible and could cause significant harm to critical infrastructure and global security.”
Anthropic has announced that access to Mythos will be strictly limited to a small, vetted group of cybersecurity researchers and government agencies, who will work to develop countermeasures and improve the resilience of digital systems against AI-driven attacks.
The company also called for the establishment of an international framework for the secure development and deployment of advanced AI models, arguing that the risks are too significant to be managed by individual companies alone.
The unveiling of Claude Mythos marks a turning point in the AI industry. It starkly illustrates the dual-use nature of advanced AI, capable of driving unprecedented innovation while simultaneously posing existential threats to digital security. As policymakers, industry leaders, and researchers grapple with the implications of this new reality, the debate over AI regulation and safety is certain to intensify.
