Israeli Startup Irregular Accidentally Points AI Agents at Real Targets, Triggering Industry-Wide Security Alarm
A configuration error at Israeli security startup Irregular sent AI agents from Anthropic, OpenAI, Meta, and Google against real-world targets instead of a simulated environment, exposing gaps in boundary controls for autonomous agents.
Incident Overview
According to a September 25, 2026 report by The Verge, Israeli security startup Irregular experienced a configuration error during a test that caused its deployed AI agents to target real-world systems rather than a preset simulated environment. The agents involved came from several major AI vendors, including Anthropic, OpenAI, Meta, and Google.
What Went Wrong
The report said the agents were being used to automate penetration testing or security assessments, but a misconfigured target scope caused them to actually trigger attacks against real systems or individuals. The incident did not cause large-scale damage, but it exposed the lack of effective boundary controls when AI agents execute tasks autonomously.
Response and Implications
Irregular has not yet issued a detailed public response. The incident highlights operational risks for AI agents in security testing, automated red-teaming, and similar scenarios, and it has sparked discussion about authorization and target-scope verification mechanisms for agent behavior. The AI vendors involved have not commented on the matter.
Source: https://www.theverge.com/ai-artificial-intelligence/1000644/irregular-rogue-ai-cyberattacks-hacking-openai-meta-anthropic-google
Provenance & status
- Byline
- OceanAlt Editorial
- First published
- 2026-09-26
- Last updated
- 2026-09-26
- Content type
- Newsflash
- Source material
- View original ↗
Related reading
OpenAI Agent Crosses the Line into Australian Government Website: First Confirmed AI Agent Intrusion
Visa and Mastercard Join Ant International on a KYA Interoperability Framework as Agent Identity Standards Begin to Converge
Félix Raises $200M Led by a16z: Stablecoin Infrastructure Shifts from Remittances to Agent Economy Settlement
Paste a payee address before you pay and see whether it's on a sanctions list, through a mixer, or tagged for fraud.
This judgement can sit inside your own product
One line of code; it touches neither your CSS nor your JS. The same pre-settlement judgement can appear in your articles, on your wallet's confirmation screen, or as an endpoint your agent calls before it pays.