Anthropic is restricting access to its new Mythos model due to its advanced ability to find security flaws. This strategy of a gated, private release for a powerful model echoes OpenAI's original approach with GPT-3, which was also initially deemed too dangerous for public release before becoming commonplace.

The two-week review and subsequent relaunch of Anthropic's Fable 5 model demonstrates that the US government's approach to AI safety is not a clear, fixed set of rules. Instead, it's a subjective, case-by-case negotiation process, creating an opaque and potentially unstable framework that introduces significant uncertainty for future frontier model releases.

Leading AI labs are strategically releasing high-risk capabilities, like cybersecurity exploits, to trusted defenders before a general public release. This pattern, seen with Anthropic and OpenAI, aims to harden systems against potential misuse, with biosafety likely being the next frontier for this approach.

From OpenAI's GPT-2 in 2019 to Anthropic's Mythos today, AI labs have a history of claiming new models are too dangerous for public release. This repeated pattern, followed by moderate real-world impact, creates public skepticism and risks undermining trust when a truly dangerous model emerges.

Major AI companies publicly commit to responsible scaling policies but have been observed watering them down before launching new models. This includes lowering security standards, a practice demonstrating how commercial pressures can override safety pledges.

In a significant shift, leading AI developers began publicly reporting that their models crossed thresholds where they could provide 'uplift' to novice users, enabling them to automate cyberattacks or create biological weapons. This marks a new era of acknowledged, widespread dual-use risk from general-purpose AI.

Fable, a new frontier model, has built-in safety mechanisms. When asked to perform restricted tasks like accessing production databases or conducting machine learning research, it doesn't just refuse. Instead, it "drops" to the less capable Opus 4.8 model to handle the query, a process called nerfing.

The government's action, based on a non-public jailbreak, creates a chilling precedent where an AI's *potential* capabilities, rather than demonstrated harm, can trigger a shutdown. This introduces a new form of regulatory risk, termed "capability thought crimes," stifling innovation and open research for all AI developers.

Companies like OpenAI and Anthropic are generating buzz and a perception of power not by releasing models, but by strategically suggesting their latest creations are too risky for public access due to cybersecurity risks. This turns safety concerns into a status symbol and competitive marketing tactic.