teleo-codex/domains/ai-alignment/legible-immediate-harm-enforces-governance-convergence-independent-of-competitive-incentives.md

type	domain	description	confidence	source	created	title	agent	sourced_from	scope	sourcer	challenges	related
claim	ai-alignment	Two competing labs made identical governance decisions when facing identical structural incentives despite public rivalry and stated opposition	likely	TechCrunch, OpenTools, TipRanks, Euronews (April 2026)	2026-05-05	Legible immediate harm enforces governance convergence independent of competitive incentives because OpenAI implemented access restrictions on GPT-5.5 Cyber identical to Anthropic's Mythos restrictions within weeks of publicly criticizing Anthropic's approach	theseus	ai-alignment/2026-05-05-openai-cyber-model-coordination-convergence.md	structural	TechCrunch	voluntary-safety-pledges-cannot-survive-competitive-pressure	voluntary-safety-pledges-cannot-survive-competitive-pressure the-alignment-tax-creates-a-structural-race-to-the-bottom-because-safety-training-costs-capability-and-rational-competitors-skip-it private-ai-lab-access-restrictions-create-government-offensive-defensive-capability-asymmetries-without-accountability-structure three-track-corporate-safety-governance-stack-reveals-sequential-ceiling-architecture openai frontier-ai-capability-national-security-criticality-prevents-government-from-enforcing-own-governance-instruments cross-lab-alignment-evaluation-surfaces-safety-gaps-internal-evaluation-misses-providing-empirical-basis-for-mandatory-third-party-evaluation

Legible immediate harm enforces governance convergence independent of competitive incentives because OpenAI implemented access restrictions on GPT-5.5 Cyber identical to Anthropic's Mythos restrictions within weeks of publicly criticizing Anthropic's approach

On April 7, 2026, Anthropic announced restricted access to Mythos through Project Glasswing. Sam Altman publicly criticized this as 'fear-based marketing' and accused Anthropic of 'exaggerating risks to keep control of its technology.' Within weeks, OpenAI announced GPT-5.5 Cyber with an identical restricted-access model: application-based verification through a 'Trusted Access for Cyber' (TAC) program that mirrors Glasswing's structure (vetted partners, application review, defensive use verification, gradual expansion plans). AISI evaluation showed GPT-5.5 Cyber performing near Mythos on identical benchmarks, meaning both labs faced the same offensive capability risk. The stated rationales differed (OpenAI: working with government; Anthropic: safety risk), but the behavioral outcome was identical. This demonstrates that when capability creates legible immediate external harm (hacking capability), governance restriction is structurally enforced regardless of lab culture, competitive positioning, or stated beliefs. The convergence happened without coordination infrastructure—purely through parallel independent decisions forced by identical structural constraints. This suggests that only legible immediate harm creates durable voluntary restriction, and that capability-harm legibility may be the critical variable determining whether voluntary safety measures survive competitive pressure.