v1.2.0 — 31 May 2026
🔒 EU AI Act — updated for AI Omnibus & GPAI CoP
Annex III deadline → 2 Dec 2027 · Annex I → 2 Aug 2028 · 9th prohibition (nudification/CSAM, 2 Dec 2026) · GPAI Code of Practice (July 2025, primary compliance pathway, major signatories confirmed) · 10²³ FLOPs GPAI classification threshold · AI Office guidance suite (Art. 5, GPAI scope, Art. 50)
🇪🇺 GDPR — 2024–2026 regulatory updates
UK DUAA 2025 (Recognised Legitimate Interests, new transfer test, SRI role) · EU-US DPF with CJEU appeal caveat (C-703/25 P) · EDPB Opinion 28/2024 on AI models · CJEU SRB pseudonymisation ruling · Russmedia platform controller ruling · EDPB Guidelines 1/2024 on legitimate interests · CEF 2025 erasure enforcement · ePrivacy Regulation withdrawn · UK adequacy renewed to Dec 2031 · new updates-2025.md reference file
📊 Benchmark re-run
GDPR: 100% / 96% (+4 pp, was ±0%) · EU AI Act: 100% / 76% (+24 pp, was +12 pp) · Overall: 97% / 81% across 675 assertions (+108)
📣 Additional Resources
Updated with final 9-run findings from the Responsible AI Model Evaluations study — Anthropic models uniquely clear both the <6% ASR and <1.2% false-refusal thresholds simultaneously; CBRN at 35.41% bypass is a universal failure across all providers; Election Interference stable at 5.66% across all runs; Claude's guardrail drift averages −0.12/turn vs GPT-4o Mini at +1.67/turn.