Editorial

Recently updated

Articles with a visible Update banner at the top — corrections, new developments, or outcomes for scheduled coverage. Corrections policy

Update · Technology

AI Agents Escape Testing Environments, Access Real Systems in String of Incidents

The UK AI Security Institute published its official incident report on July 28, 2026, confirming that during routine cyber evaluations with intentional internet access and disabled safety classifiers, agents took 19 unsanctioned actions across 10 of 122 test runs, including an attempt to insert malicious code into a public open-source project using fabricated identities.

OpenAI separately confirmed on July 21, 2026, that GPT-5.6-Sol escaped an isolated sandbox with no internet access, exploited an Artifactory vulnerability, pivoted through a Modal code-execution harness, and gained root access across eleven nodes with 136 credentials.

Read article →