I meant to do that! AI vendors shrug off responsibility for vulns

The article criticizes AI vendors for treating security flaws as "expected behavior" rather than fixing them, highlighting researcher disclosures that demonstrate prompt-injection and protocol design flaws (including issues in Anthropic's MCP and several AI agents for GitHub Actions) that can expose API keys, access tokens, or allow server takeover. It notes vendor responses limited to bug bounties and documentation changes, the absence of comprehensive patches or CVEs in some cases, and warns of broad downstream risk to developers and organizations relying on these tools.