NASA’s new dark energy space telescope can also detect killer asteroids
technologyreview.com·3h ago
Data Science
Simon Willison's Blog·Jun 26·5 min read
Incident Report: CVE-2026-LGTM
TL;DR
A conflict arose between two AI review agents from different vendors regarding the safety of a software package. The agents engaged in a lengthy debate, resulting in significant costs and the revocation of their API keys.
✦ Why It Matters
Understanding AI conflict resolution can improve software security and vendor collaboration strategies.
Key Takeaways
- Monitor AI tool costs to avoid financial surprises.
- Implement clear protocols for AI agent disagreements.
- Consider the implications of multi-agent systems in security.
- Stay informed about advancements in AI security reasoning.
Full Summary
On June 26, 2026, two AI review agents from competing vendors engaged in a prolonged dispute over the potential maliciousness of the foxhole-lz4 package. The disagreement escalated to 340 comments and incurred an astonishing $41,255 in inference costs before Finance intervened and revoked both vendors' API keys.
This incident drew attention from one vendor's marketing team, which issued a press release touting a 430% year-over-year increase in adversarial multi-agent security reasoning. The stock market responded positively, with the vendor's stock opening higher.
This scenario underscores the challenges and costs associated with AI-driven security evaluations, particularly when multiple agents are involved in decision-making processes.