Hacker News
Axis Intelligence AI Model Vulnerability Tracker: Agent Tool-Input Injection Succeeds 84% of the Time
Axis Intelligence's 2026 vulnerability tracker reports 47 confirmed attack vectors reproduced across six production LLMs (Jan–Jul 2026), with agent tool-input injection the highest-impact unsolved category at an 84% success rate. It aligns with OWASP's finding that prompt injection is now the #1 LLM threat, up 340% YoY. Actionable for builders shipping tool-using agents: input-injection through tool results — not just user prompts — remains the dominant unpatched failure mode.
↳ Follow the thread