Biosecurity Safeguards

Biosecurity safeguards are measures AI developers and deployers use to stop their models from giving meaningful uplift to people trying to create or spread biological weapons or dangerous pathogens. For governance teams, this covers things like refusal policies on biology queries, restricted access to the most capable models, red-team testing by outside experts, and internal review thresholds that trigger when a model's capabilities approach a defined risk level. Regulators and frontier AI developers increasingly treat these controls as a core obligation, so the reliability of these safeguards is a question boards and legal teams need to be able to answer.

1 item

Corporate PolicyGlobal2026-11-10

Anthropic's Fable 5.2 and Mythos 5.2 Create a Two-Tier Access Control Problem

Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on November 9, 2026, two variants of the same frontier model optimized for advanced technical work. Fable 5.2 is available generally with enhanced cybersecurity and biosecurity restrictions. Mythos 5.2 carries fewer constraints and is restricted to organizations that pass Anthropic's Life Sciences and Cyber Verification Programs. Compliance teams must now account for which variant their organization or vendors can access, and whether internal controls justify that access tier.

Anthropic Claude Fable 5.2 Mythos 5.2 frontier model release biosecurity safeguards access controls vendor due diligence dual-use AI