Anthropic Releases Claude Mythos 5 AI Model to Public with Security Restrictions
Anthropic released Fable 5, a restricted version of its advanced Claude Mythos AI model, to the general public on Tuesday, while keeping the unrestricted Claude Mythos 5 available only to select partner organizations. The company implemented safeguards to prevent the model from answering queries about cybersecurity vulnerabilities, biology, and chemistry due to concerns about potential misuse in critical infrastructure attacks. The release reflects ongoing tensions between AI safety and accessibility, with the model priced at double the cost of its predecessor.
Anthropic made its advanced Claude Mythos AI model partially available to the public through Fable 5, a restricted version designed for writing code, research analysis, and image processing. The company simultaneously deployed the unrestricted Claude Mythos 5 to approximately 200 partner organizations in over 15 countries, including cybersecurity firms and government agencies. Anthropic implemented multiple safeguards, routing sensitive queries about cybersecurity, biology, and chemistry to the less capable Opus 4.8 model, and blocking attempts to extract the technology for use in authoritarian countries. The company conducted extensive red-teaming and bug bounty programs involving over 1,000 hours of security testing to identify potential bypasses. Fable 5 carries a premium price of $10 per million input tokens and $50 per million output tokens—double the cost of Opus 4.8—reflecting the company's substantial computing infrastructure expenses.
What's missing
The articles lack clarity on what specific vulnerabilities Mythos can identify in critical infrastructure and whether independent security researchers have verified these claims. Additionally, there is limited discussion of how effective the routing safeguards actually are in practice or whether users can circumvent them through prompt engineering.
How coverage differed
The Guardian's coverage emphasizes Anthropic's disputes with the Trump administration over surveillance and lethal weapons restrictions, framing the company as principled but facing consequences. The article also highlights financial pressures and IPO plans, potentially suggesting the release is partly motivated by commercial interests rather than purely safety considerations.
What different sources said
Anthropic releases ‘safe’ version of Claude Mythos AI model to public
Related
Taiwan Considers Stricter AI Chip Export Controls to China to Align with U.S. Restrictions
Taiwan is considering implementing stricter export controls on AI chip sales to China to better align with U.S. semiconductor restrictions and combat smuggling. Currently, Taiwan lacks specific laws treating unauthorized AI chip exports to China as crimes, relying instead on enforcement through other existing regulations. This move matters because it would close legal gaps that allow advanced semiconductor diversion and strengthen the U.S.-Taiwan-led effort to prevent China from accessing cutting-edge AI technology for military purposes.
Microsoft Patches High-Severity Zero-Days Disclosed by Researcher in Ongoing Dispute
Microsoft released fixes for two high-severity zero-day vulnerabilities that were publicly disclosed by a researcher known as Nightmare Eclipse. The researcher claims Microsoft violated an agreement regarding vulnerability handling, leading to the public disclosure with proof-of-concept code. The incident highlights tensions between security researchers and major tech companies over responsible disclosure practices.
Commonwealth Fusion Systems Publishes Physics Research Supporting 400 MW Reactor Design
Commonwealth Fusion Systems released five peer-reviewed papers detailing the physics basis for its ARC fusion reactor design, which would generate 400 MW of power. The company is pursuing a faster timeline than the international ITER project by using high-temperature superconductors to build smaller, more efficient tokamak reactors. The research represents an important step in validating whether private fusion companies can achieve commercial viability before solar and other renewable technologies dominate the energy market.