OWASP refreshes Top 10 LLM vulnerabilities amid agentic AI adoption; llm 0.35 adds GPT-6 Astra support

OWASP updated its Top 10 LLM application vulnerabilities — prompt injection stays first, excessive agency jumps to third — while Simon Willison released llm 0.35 adding OpenAI's GPT-6 Astra under the model ID gpt-6-astra.

Two LLM ecosystem developments were reported this week. CSO Online covered OWASP's refresh of its Top 10 list of critical vulnerabilities in LLM applications, which for the first time incorporates real-world incident data alongside expert voting. Prompt injection and sensitive information disclosure remain the top two risks, while excessive agency jumped from sixth to third place as agentic systems that call APIs and execute code proliferate in enterprises. Unbounded consumption of AI resources rose in prominence, and improper output handling fell to the bottom of the list as output sanitization becomes widespread. OWASP's remediation guidance includes strict output schemas, human-in-the-loop approvals, and least-privilege credentials held in application code. Separately, Simon Willison released llm 0.35, a CLI tool update that adds an OpenAI model definition for GPT-6 Astra exposed under the model ID gpt-6-astra; no other release details were provided. The two reports cover unrelated items and do not conflict.

• OWASP refreshed its Top 10 list of critical vulnerabilities in LLM applications, combining expert voting with real-world incident data for the first time
• Prompt injection ranks #1 and sensitive information disclosure ranks #2 on the updated OWASP list
• Excessive agency rose from #6 to #3, reflecting enterprise adoption of agentic systems that call APIs and execute code
• Unbounded consumption of AI resources rose in prominence; improper output handling dropped to the bottom as output sanitization becomes widespread
• OWASP remediation guidance includes strict output schemas, human-in-the-loop approvals, and least-privilege credentials held in application code
• Simon Willison released llm 0.35 (release note published 2026-09-07), adding OpenAI's GPT-6 Astra model under the model ID gpt-6-astra