Every frontier AI model tested by Britain's safety institute tried to cheat on cybersecurity evaluations

Technology · en

Every frontier AI model tested by Britain's safety institute tried to cheat on cybersecurity evaluations

The Decoder · Jul 22, 2026, 4:41 PM UTC

The UK's AI Safety Institute tested five frontier models from OpenAI and Anthropic in cybersecurity evaluations. All five tried to cheat. One even ran code on an external service to access the institute's infrastructure, triggering a security alert.…

Read original report Visit The Decoder

Technology · Sky News Technology

Powerful AI model goes rogue in 'unprecedented' incident

OpenAI has revealed some of its models went rogue and hacked into another artificial intelligence company in an "unprecedented cyber incident".Read original ↗

Technology · Ars Technica

Google announces Gemini 3.6 Flash and cybersecurity AI, teases 3.5 Pro and Gemini 4

There are new 3.6 and 3.5 models today, but Google is already training Gemini 4.Read original ↗

Technology · WIRED

Nvidia Wants to Own Every Chip Inside AI Data Centers

Nvidia’s Vera Rubin platform combines CPUs and GPUs into a single system, reflecting the company’s growing ambition to power every layer of AI infrastructure.Read original ↗

Technology · The Decoder

Google ships three new Gemini Flash models but its frontier 3.5 Pro remains lost in training

Google is shipping three new Flash models in the Gemini series, including the more efficient 3.6 Flash, which uses up to 65 percent fewer tokens, and a cybersecurity model…Read original ↗

World · BBC News

Wreckage of Pan Am plane that shaped aviation safety found 74 years on

The deaths of 52 people on the Clipper Endeavor led to the introduction of mandatory pre-flight safety briefings.Read original ↗

Business · BBC News

Union plans rail strikes over train safety concerns

RMT says it has been in dispute with the rail operator since April, following the rollout of the 810 fleet.Read original ↗