Trump says Iran war "close to over" amid hopes for more negotiations
In Claude news today, the UK’s AI Safety Institute has confirmed that Anthropic’s Claude Mythos Preview can autonomously execute sophisticated multi-stage cyberattacks at success rates no prior AI model has approached, completing a 32-step corporate network intrusion simulation that had never been finished by any AI system.
These findings reframe AI safety assessments from theoretical benchmarks into operational risk disclosures, with direct implications for enterprise deployment decisions and the financial sector’s security posture.
Anthropic confirmed the model’s existence on April 13, 2026, weeks after its presence was first surfaced via a late-March website leak, and announced it would not release Claude Mythos Preview publicly, citing the model’s autonomous offensive capability rather than regulatory or safety-threshold constraints.
US Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell reportedly warned bank executives about the threat posed by the model in the days preceding the institute’s formal disclosure.
Claude News: Anthropic Mythos Review: What the Safety Assessment Actually Found
The UK AI Safety Institute recently evaluated Mythos Preview against The Last Ones (TLO), a complex corporate network attack simulation. Mythos Preview succeeded in 3 out of 10 attempts, completing an average of 22 out of 32 steps, outperforming Claude Opus 4.6, which averaged only 16 steps.
The evaluation required significant computational resources, with sessions consuming up to 100 million tokens, and performance improved with additional computing power.
Beyond simulation tasks, Mythos Preview autonomously identified thousands of zero-day vulnerabilities across major operating systems, including long-standing flaws that had remained undetected for years.
Anthropic’s internal tests found that engineers could direct the model to find remote code execution flaws overnight, yielding complete exploits by morning. In tests targeting privilege escalation, over half of the 40 vulnerabilities resulted in successful exploit chains without human intervention.
However, experts have cautioned against framing Mythos as a “super-hacker,” noting that while its capabilities are real, the confirmed number of severe vulnerabilities is much smaller, emphasizing the importance of accurate risk assessment in enterprise AI governance.
Generative AI Stack: How Safety Scrutiny Affects Valuations Across the Sector
SOURCE: StockAnalysis
The Claude news came as Anthropic, a private company, is now valued at about $61.5Bn, offering no direct equity exposure for public investors. Its valuation influences public companies’ AI credibility.
Palantir Technologies, a competitor for AI contracts, has faced selloff pressure due to AI safety concerns, but the delay in the release of Claude Mythos eliminates a competitive threat. However, increased scrutiny raises regulatory hurdles for all AI vendors, including Palantir’s AIP platform.
In contrast, CrowdStrike Holdings and Palo Alto Networks, as partners in Project Glasswing, stand to gain from enhanced vulnerability detection through Mythos.
Companies like Amazon and Google, which support Anthropic’s computing needs, will see neutral to positive effects from the Mythos disclosure, indicating ongoing compute demand. Investors must consider the impact of evolving AI regulations on stock valuations as formal oversight is expected to develop by 2026.
What Investors Should Watch as AI Safety Oversight Evolves
Monitor the patch velocity for vulnerabilities identified by Mythos Preview, as Anthropic reports over 99% remain unpatched.
The timeline for coordinated disclosure with Project Glasswing partners will influence whether the narrative remains focused on defense or shifts to an AI-enabled security incident if an exploit occurs.
Any incident involving a Glasswing partner could impact the stock prices of CRWD, PANW, or the affected operator.
On the regulatory side, look for comments from the EU AI Office regarding whether Mythos Preview’s capabilities require mandatory conformity assessments under the AI Act, which could set a precedent for frontier cybersecurity models.
The upcoming earnings reports for AMZN, GOOGL, and MSFT in late April and early May will be key for insights into the economics of the Glasswing partnership and Mythos-related compute demand.
However, a critical uncertainty remains how quickly hostile state actors can reach the capability threshold demonstrated by Mythos Preview with this Claude News, an issue that will not be resolved before the next annual capability review in 2027.
***
Looking to start your trading day ahead of the curve?
Get up to speed before the bell with Bull Whisper—a sharp, daily premarket newsletter packed with key news, market-moving updates, and actionable insights for traders.
Start your day with an edge. Subscribe to Bull Whisper using this link.