The ChatGPT maker claims covert influence operations targeted an American debate, but says it found no meaningful impact.

Two separate PRC playbooks collide: a revived botnet cluster and banned AI-driven social campaigns that tried to steer an AI datacenter debate.

A new model architecture swaps brute-force token prediction for instruction-response training, cutting cost and compute.

A suspected influence operation tried to sour opinion online, but OpenAI says it never broke out meaningfully.

A Gmail-connected OpenClaw agent failed to verify the requester, proving phishing can beat AI access controls.

Record lender credit activity and tougher-to-get pricing show up in JLL's April data, with knock-on effects for borrowers and investors.