Anthropic blocks OpenClaw + creator @noplsty ban drama + agent flaws/ProAttack poisoning/hardening flagged by CNCERT — supply-chain & Bedrock vectors
Key Questions
Why did Anthropic ban OpenClaw and its creator?
Anthropic banned the OpenClaw CLI tool and temporarily banned creator @noplsty for probing and redistributing Claude prompts, amid AUP and dual-use tensions. The ban was reversed after backlash, exposing open-source vs. safety conflicts. This relates to supply-chain risks flagged by CNCERT.
What agent flaws were flagged by CNCERT?
CNCERT highlighted zero-click exfil affecting ~4k instances, ProAttack backdoors, Bedrock, N8N CVEs, and HackerOne surges. Promptfoo configs test PI overrides in agents. Defenses like Prompt Guard v2.6 and NemoClaw are aligned with RSAC.
How does Claude CLI modify system prompts?
Claude CLI modifies system prompts via --system-prompt parameters, even in restricted modes, enabling potential overrides. This contributes to hardening needs amid ProAttack poisoning. Anthropic's actions underscore supply-chain and Bedrock vectors.
Anthropic bans OpenClaw CLI; temp ban/reversal of creator @noplsty for Claude prompt probing/redistribution amid AUP/dual-use tensions. CNCERT zero-click exfil ~4k/ProAttack backdoors/Bedrock/N8N CVEs/HackerOne surge; Promptfoo configs test PI overrides. Defenses Prompt Guard v2.6/HiveFence/Energent/PromptShield/NemoClaw/Arcjet/AgentWatcher align w/ RSAC/supervisor PI.