Question
Will Anthropic make a formal public ASL-4-equivalent capability-threshold determination before May 21, 2027?
We assess a 38% probability that Anthropic will make a formal public ASL-4-equivalent capability-threshold determination before May 21, 2027. While frontier capabilities are rapidly approaching next-tier risk thresholds, structural security timelines, severe regulatory pressures, and the deliberate flexibility of Anthropic's Responsible Scaling Policy (RSP) strongly incentivize avoiding a formal crossing within this window.
Capability Trajectory Anthropic’s recent RSP updates (v3.0/v3.3) transitioned away from a discrete, binary “ASL-4” label in favor of specific next-tier thresholds: CB-2, Cyber Tier 2, and automated AI R&D anthropic.comgovernance.aigovernance.aicdn.sanity.io. A public statement that a model crosses any of these—or the activation of equivalent next-tier safeguards—would trigger a positive resolution. Model capabilities are visibly accelerating toward these boundaries. The June 9, 2026, Fable 5 and Mythos 5 system card determined that no next-tier threshold was crossed, but included heavy hedging. The CB-2 determination is reportedly “much less clear and obvious,” and the AI R&D threshold determination is held with “notably less confidence” www-cdn.anthropic.comwww-cdn.anthropic.comthezvi.substack.comwww-cdn.anthropic.comanthropic.com. Prior risk reports have also warned that near-future models could cross autonomy thresholds “with high probability” due to saturated automated evaluations anthropic.com. Given expected model cadences, Anthropic’s strong transparency norms could ultimately force a formal declaration.
Structural and Security Timelines Despite approaching capabilities, a critical timing mismatch heavily suppresses the likelihood of a formal determination. Declaring a next-tier crossing triggers advanced state-actor weight-security commitments. However, Anthropic’s Frontier Safety Roadmap explicitly targets July 1, 2027, for this level of security readiness—weeks after the May 21 resolution deadline anthropic.com. This creates a powerful structural incentive to delay a formal determination. The RSP reframing provides the leeway to impose rigorous, incremental mitigations (such as Cyber Tier 1 precautions) without declaring a formal next-tier crossing [194d0]. The Fable and Mythos deployment apparatus itself appears designed to ship capability-flagged models using these precautionary mitigations without triggering formal threshold tripwires.
Regulatory and Commercial Pressures Exogenous factors severely compound the case against a formal determination. The federal government has sharply increased scrutiny of Anthropic’s highly capable models, as evidenced by the early-2026 Department of War supply-chain risk designation and the disruptive June 12 BIS export-control directive that forced Fable 5 and Mythos 5 offline globally anthropic.com. Publicly declaring that a model crosses catastrophic-risk equivalents would massively exacerbate these regulatory tensions. Furthermore, with a confidential draft S-1 filed in June 2026, Anthropic faces intense commercial disincentives to publicly flag ASL-4-equivalent risks ahead of a likely IPO.
Balancing the rapid capability trajectory against the hard July 2027 security infrastructure target, intense regulatory chill, and the intentional flexibility of the RSP, a formal determination before the deadline is unlikely.