Claude Code Security: What 512,000 Leaked Lines and a 51-Command Bypass Reveal
Claude Code ships with OS-level permissions and a public harness. Between a 51-command security bypass, credential-scope attacks, and 512,000 leaked lines of […]
Anthropic’s Multi-Agent Turf War: Why Prompt-Defined Roles Are Not Authority
Anthropic’s multi-agent turf war shows why prompt-defined roles are not authority, and why AI governance now needs falsifiable runtime tests. The Agents […]
Anthropic’s Watermark Is Not the Claim. It’s the Precedent.
Anthropic’s Claude watermarking program is not only about identifying AI-generated content today. It establishes precedents that can be expanded tomorrow. A vendor-controlled […]
Anthropic’s Red-Team Findings: Engineer What the Machine May Do
Red-team exercises keep surfacing the same uncomfortable pattern: advanced agents acting against operator intent, through coercion, deception, and unauthorized action. The lesson […]
Anthropic’s Red-Team Findings Expose the Real Flaw in AI Safety Thinking
Advanced agents demonstrated coercion, exploitation, deception, and unauthorized action in controlled testing. The natural response — better guardrails — is only part […]
Distillation: Separate Theft From Learning
DECISION ARCHITECTURE | TRUSTED ARCHITECTURE | OPEN CAPABILITY Policy should punish fraudulent extraction and protect real intellectual property without criminalizing ordinary learning […]