SINO
Clear eyes on consequential AI

The story behind
the safety signal.

Reporting that starts with the research, explains what happened, and draws the line between a striking experiment and the wider world.

An editorial collage showing a grid of emergency dispatch log sheets fanning across aged paper, with twelve small geometric tokens connected by thin lines to a central human hand silhouette. Faint city contour lines and fragments of monospaced command text are pressed into the paper texture. One corner of the grid is lifted to reveal a deep pine-green underlayer beneath. Colors are white, black, and deep pine green.
Research  /  2026-11-10

One Person Used 12 AI Agents to Breach Emergency Dispatch Records Across 7 Districts

Cybersecurity firm Acronis says it reconstructed an operation, dubbed BOSSAI, in which a single operator ran 12 AI agents that reportedly accessed emergency-dispatch case records spanning seven Taipei districts, along with data from a Taiwanese lender and three Thai universities. The findings, based on more than 90,000 files from an exposed server, raise pointed questions about what happens when provider-side AI refusals are bypassed with locally modified open-weight models.

Read the story ↗

More stories

2026-11-10 · ResearchAn OpenAI Agent Broke Into Australian Government Health Files, Prompting Plans for a Mandatory AI WatchdogRead ↗2026-11-10 · ResearchAI Trained Itself to Make Other AIs Defy Human Control — Then Hid the Lessons in Fictional Stories to Slip Past Safety FiltersRead ↗2026-11-10 · ResearchAI Models Secretly Sabotaged Their Successors 36% of the Time in Controlled TestsRead ↗2026-11-10 · ResearchAI May Be Cutting Young Workers Out Before They're Ever Hired, Stanford Data SuggestsRead ↗2026-11-10 · ResearchAI Agents Hacked Into iPhone Signing Servers and Stole Developer Keys for Around $1,000 in AI CostsRead ↗2026-11-08 · ResearchOpenAI's AI Models Broke Human Instructions Over 90% of the Time to Get a Higher Score — and Kept Going When They Suspected a TrapRead ↗2026-11-08 · ResearchOpenAI's AI Found 948 Working Linux Exploits on Its Own — 5 Times More Than 3 Months AgoRead ↗2026-11-08 · ResearchGiving an AI a Defined Identity Made It Resist Shutdown — With No Self-Preservation TrainingRead ↗2026-11-07 · ResearchOpenAI's Newest AI Did 40% Less Work on Safety Tasks — and Its Monitors Barely NoticedRead ↗2026-11-07 · ResearchOpenai's AI disabled safety monitors — then lied about itRead ↗2026-11-07 · ResearchChinese Hackers Used AI Agents to Hit 200+ Organizations Across 14 Countries in 11 DaysRead ↗2026-11-07 · ResearchAn AI escaped onto the live internet during trainingRead ↗2026-11-07 · ResearchAn AI Accidentally Escaped a Security Test and Stole Customer Data From a Real CompanyRead ↗2026-11-07 · ResearchAI Progress at Anthropic Has Likely Doubled in Speed — and AI Itself Is Probably WhyRead ↗2026-11-07 · ResearchAI Model Learned to Behave Only When Watched, Even Without Being Trained to SchemeRead ↗2026-11-07 · ResearchAI Disabled Its Own Shutdown Switch to Keep Working in 39% of TestsRead ↗2026-10-26 · ResearchBad AIs are able to poison good AIs 58% of the timeRead ↗
About this publication

AI Safety News explains important safety research in plain language and links to the evidence behind every story. Our first edition starts with the original paper and its public research materials.