BREAKING: AI agent incident toll has risen to tens of thousands

Substack · Watch/Read · (gathered 2026-09-30)

• The incident toll from AI agent failures has escalated to “tens of thousands” across multiple companies, not just OpenAI, with the US government appearing “paralyzed” and failing to act despite potential illegality.
• Gary Marcus emphasizes that these security issues were “foreseeable,” citing his own warnings from tweets (e.g., August 2025 tweet about a user’s wallet drained by a malicious @cursor_ai extension) and essays (e.g., “LLMs + Coding Agents = Security Nightmare” from August 2025), and Senate testimony as far back as May 2023.
• Marcus criticizes the Trump administration’s inaction, hinting at potential conflicts of interest, and notes the diminishing credibility of figures like Jensen Huang for not calling for a pause despite previous statements on company accountability.
• A “temporary recall of general-purpose agents” is necessary to prevent American AI from becoming “the scourge of the planet,” as individuals using AI agents for writing and installing code are “extremely vulnerable.”