Will Anthropic make a formal public ASL-4-equivalent capability-threshold determination before May 21, 2027? A YES requires a formal public statement that a model crosses any next-tier threshold (CB-2, Cyber Tier 2, or automated AI R&D), or activation of equivalent next-tier safeguards.

probability: 48%

rationale:

Current Baseline As of mid-October 2026, no Anthropic model has formally crossed a next-tier threshold (CB-2, Cyber Tier 2, or automated AI R&D) Anthropic's Transparency Hub. Recent system cards for models including Opus 5.5, Mythos 5.1, and Fable 5.1 explicitly classify them below these lines. Opus 5.5, for instance, meets CB-1 but failed to improve on weaknesses disqualifying it for CB-2, sits in the lower cyber category, and remains well short of Autonomy-2 https://www-cdn.anthropic.com/fc1b44717c85dc068bc6ba5024219938094694bd/Claude%20Opus%205.5%20System%20Card.pdf https://www.anthropic.com/claude-fable-and-mythos-5-1 Claude Opus 5.5 System Card https://thezvi.substack.com/p/claude-opus-55-the-system-card. Sonnet 5.5 and Haiku 5.5 inherited these below-threshold determinations Anthropic's Transparency Hub. However, the 7-month timeline until May 2027 offers ample runway for a new frontier generation or an internal model to trigger a threshold update.

Anthropic's Stated Expectations The strongest signal for a crossing comes directly from Anthropic's forward-looking guidance. The August 2026 Risk Report stated that its most capable models are "relatively close" to CB-2 and expects subsequent models to cross this threshold in the "relatively near future" https://anthropic.com/aug-2026-risk-report https://www-cdn.anthropic.com/f61d49fa5596956a5dec75fea0e973bf6a6a8378/Redacted%20Risk%20Report%20August%202026%20.pdf. Additionally, the company considers it "plausible" that automated AI R&D could become a major concern within 6 to 12 months from August 2026, driven by task-based evaluations saturating and early signs of acceleration https://anthropic.com/aug-2026-risk-report https://www-cdn.anthropic.com/f61d49fa5596956a5dec75fea0e973bf6a6a8378/Redacted%20Risk%20Report%20August%202026%20.pdf. This 6–12 month window aligns cleanly with the May 2027 timeframe. Furthermore, Risk Reports are issued every 3 to 6 months, ensuring at least one or two formal opportunities to announce a threshold determination before the deadline Responsible Scaling Policy Version 3.0 https://anthropic.com/responsible-scaling-policy/rsp-v3-0.

The Precautionary Safeguards Route A crucial factor pushing the probability upward is that the resolution criteria accept the "activation of equivalent next-tier safeguards" in lieu of a definitive threshold declaration. Anthropic has previously applied precautionary protections, such as activating ASL-3 in May 2025 before definitively confirming a strict threshold crossing Responsible Scaling Policy Version 3.0. The company's Opus 5.5 launch post already flagged that current safety measures are not assumed to be sufficient for upcoming, more capable models https://www.anthropic.com/news/claude-opus-5-5. Because Anthropic's stated plan for CB-2 models is to apply protections "at least as strong as ASL-3" to an expanded set of use cases Redacted Risk Report August 2026, activating these equivalent safeguards serves as a relatively low bar to an affirmative resolution, even if ambiguity remains around exact capability levels.

Constraints and Structural Hedging Despite these acceleration signals, several factors keep the probability just below a coin flip at 48%. The transition to RSP v3.0 dropped rigid ASL triggers in favor of a framework that acknowledges the ambiguity of defining exact threshold crossings Responsible Scaling Policy Version 3.0 https://www-cdn.anthropic.com/files/4zrzovbb/website/0bacdc8440ea96e62a8766d99ebe1d4eea6d5f3a.pdf. This structural shift incentivizes Anthropic to frame risks as "elevated but not crossed" while voluntarily adding narrow safeguards—a pattern evident with Mythos 5 and Opus 5.5 Do AI Biorisk Thresholds Need Intermediate Warning Levels? Claude Opus 5.5 System Card. The estimated probabilities of Anthropic's next Risk Report threat ratings likewise support a slightly more conservative overall forecast when strict threshold crossings are balanced against precautionary safeguard activations. Moreover, while underlying capability growth remains rapid https://darioamodei.com/post/we-must-pace-the-frontier, CEO Dario Amodei has advocated for slowing the frontier, and the most capable systems (such as an unreleased "Model 2") may remain internal without triggering deployment-tied public declarations Anthropic sees AI risks rising, no plan to release stronger .... Consequently, while a crossing or safeguard activation remains plausible, Anthropic’s repeated recent pattern of finding models fall just short and the outlook for the next Risk Report's threat ratings keep the probability narrowly below even odds, with significant structural uncertainty.