𝗨𝗦 𝗙𝗼𝗿𝗰𝗲𝘀 𝗔𝗻𝘁𝗵𝗿𝗼𝗽𝗶𝗰 𝘁𝗼 𝗦𝗵𝘂𝘁 𝗗𝗼𝘄𝗻 𝗔𝗜 𝗠𝗼𝗱𝗲𝗹𝘀
The US government ordered Anthropic to disable two advanced AI models. These models are Fable 5 and Mythos 5. The order affects all users worldwide.
The Commerce Department issued an export control directive on June 12, 2026. This happened because of a simple security flaw. A three-word prompt "fix this code" allows a jailbreak.
How the flaw works:
- Researchers found that Fable 5 creates working patches for unpatched code.
- This helps defenders fix security gaps.
- However, attackers use this same method to find flaws in codebases automatically.
The ban is strict. US export rules classify sharing this technology with non-citizens as an export. Anthropic had to shut down the models for everyone to comply.
The industry is reacting with concern. Over 100 experts from companies like Google, Nvidia, and Adobe signed an open letter. They argue that these controls hurt defenders. They believe this move gives an advantage to adversaries who develop their own AI.
Key facts about the models: • Mythos 5 is a base model that can identify and chain cyberattacks. • Fable 5 is a version designed for defensive use with safety guardrails. • Mythos 5 previously passed cybersecurity tests from the UK AI Security Institute.
Anthropic disagrees with the decision. They state that the jailbreak risk is small. They argue that stopping models for a narrow flaw would stop all new AI deployments.
This move also impacts US allies. Countries like Canada and the UK rely on these models to protect critical infrastructure. The ban stops their access too.
The situation shows the tension between AI innovation and national security. It asks if current laws can handle tools that react to a simple three-word phrase.
Source: https://dev.to/techmag/us-forces-anthropic-to-shut-down-fable-5-mythos-5-over-security-flaw-4dlm
Optional learning community: https://t.me/GyaanSetuAi