BREAKING
New 'Agent Data Injection' Attack
Six Systems Affected
Coding agents
Claude Code
OpenAI Codex CLI
Google Gemini CLI
Browser agents
Claude in Chrome
Antigravity
Nanobrowser
How ADI Fools Agents
1Plant fake content
2Agent trusts it
3Wrong click or RCE
Attack Success Rates
JSON inject43.3
Web DOM100
Under defenses50
0%
skills vulnerable
0%
malicious intent
0
skills analyzed
Approval Prompts Aren't Enough
AI NEWS BLITZ
Researchers disclose a new attack class that tricks AI agents into trusting forged data.