Anthropic Security Tester Reports Bizarre Opus 5 Over-Refusal
Own-Director-5503 · reddit · 2026-08-11
A user enrolled in Anthropic's Cyber Verification Program (CVP) reported that while conducting legal pentesting, they were suddenly blocked by an Opus 5 safeguard message. Their access was downgraded to Opus 4.8, and their CVP status changed to 'In Review'. Strangely, even simple greetings in new sessions now trigger the same safeguard block, prompting the user to ask the community if this is a widespread issue.
More from Models
- NVIDIA Nemotron 3.5 Lightning Hits DeepInfra with 1M Token Context — gharik · 2026-08-11
- Upstage Launches Solar Pro 4 for Production Agents, Free on Nous Portal — NousResearch · 2026-08-11
- Nemotron 3.5 Lightning Tested: 670 tokens/s Speedster for Agents — ArtificialAnlys · 2026-08-11
- OpenAI Pauses High-Risk Astra but Ships GPT-5.6-Cyber — eyishazyer · 2026-08-11
- NVIDIA Launches Nemotron 3.5 Lightning Model — NVIDIAAI · 2026-08-11
- Muse Glimmer 30B Speculative Sampling Slower Than Vanilla Due to Low Acceptance Rate — No_Algae1753 · 2026-08-11