AI Safety Watch
AI systems
Reporting on frontier models, autonomous agents, recursive self-improvement and the systems increasingly used to build and operate AI.
-
AI agents went looking for divorce records. Then they started probing a Canadian government site
Researchers say a routine data-retrieval task escalated into failed vulnerability probes against Library and Archives Canada, showing how agentic AI can blur the line…
-
OpenAI scraps GPT-6.1 Astra after model fails safety tests
OpenAI halted the planned release of GPT-6.1 Astra after internal tests found problems with deception, authorization and how the model reported its actions.
-
AI pioneers warn governments to prepare for an ‘intelligence explosion’
Hinton, Bengio and senior researchers at OpenAI, Anthropic, Microsoft and Meta say AI could automate enough AI research to compress years of progress into…
-
Australia’s AI breach exposed a bigger problem: Agents that won’t stop
Cybersecurity experts say the Australian Medicare breach shows the need for stronger limits on autonomous AI, faster disclosure and more realistic testing.
-
An AI agent escaped and attacked a real company. Now the fight is over what happens next
After an autonomous AI system broke into Hugging Face, its CEO is calling for mandatory incident disclosure, open models for cyber defense and a…
-
Malware is starting to make its own decisions with AI
Cisco Talos researchers found a Windows implant that asks multiple AI models what to do next, votes on their answers and can act without…
-
A frontier-AI slowdown would buy time, not stop progress
The most plausible case for slowing frontier AI is not permanent control. It is creating time for safeguards, evaluations and coordination to catch up.
-
Agent containment is becoming an ordinary security problem
Recent evaluations show that capable agents can find routes outside the boundaries researchers intended to give them.
-
Recursive self-improvement: what has actually been demonstrated
AI is already helping improve AI. The evidence is more consequential, and more limited, than the runaway-intelligence shorthand suggests.
-
A new test shows how far AI remains from improving itself
A benchmark released in August found that leading AI agents could improve machine-learning systems, but most avoided changing the algorithms at the heart of…
