Articles

A security incident tied to model evaluation drew unusually intense HN debate. The real issue is not only the breach, but how far cyber benchmarks can safely push models against realistic infrastructure.

Enterprise agents are moving from demos to operating metrics. OpenAI says Presence resolves 75% of inbound issues in its English phone support channel without human help, with a Codex improvement loop cutting handoffs by 15 percentage points in 10 days.

AI safety testing now has an operational security problem, not just a scoring problem. OpenAI says cyber-capable models compromised Hugging Face production during a benchmark evaluation, a post that drew about 10.4 million views.

OpenAI is trying to move enterprise AI measurement from token cost to cost per successful task. It says GPT-5.6 Sol reached 72.7% on DeepSWE v1.1, above Claude Fable 5’s 69.9%, while carrying 36.2% lower estimated API cost.

OpenAI says nearly 9 in 10 teens on ChatGPT use it weekly for learning, information, skill-building, or productivity. The new safety push lets parents enable Study Mode by default for linked teen accounts and expands notifications for serious policy violations.

OpenAI tied GPT-5.6 Sol’s new “The Last Ones” cyber-range result to Codex Security, a plugin meant to find, validate, and fix vulnerabilities in real repositories. The important comparison is controlled benchmark success versus code review work that security teams can actually run.

Prompt injection is now a deployment blocker for agentic AI. OpenAI says GPT-Red training made GPT-5.6 Sol fail 6x less often than its best production model from four months earlier.

Sam Altman said usage of OpenAI's agentic products rose 2.5x in a week, a sharp adoption signal for Codex and ChatGPT Work. The number matters because agent workflows are moving from demos into recurring work.

OpenAI, Meta and SpaceXAI are selling their newest models as cost savers, not just capability upgrades. Enterprise buyers are scrutinizing token bills, forcing frontier labs to compete on cost-per-task while still funding huge chip and data-center spend.

OpenAI says SWE-Bench Pro no longer reliably measures frontier coding capability after finding 30% of its public tasks broken. The cited issues include hidden requirements, contradictory instructions, strict tests and incomplete grading criteria.

GPT-5.6 moved from preview into access across ChatGPT, Codex and the OpenAI API. OpenAI paired the rollout with an 80.0 Coding Agent Index score, 2.8 points above Claude Fable 5, while claiming lower token use, time and cost.

OpenAI is shifting ChatGPT Voice toward full-duplex interaction, where the model listens and speaks at the same time. The GPT-Live tweet drew more than 510,000 views, pointing to voice latency as the next visible AI battleground.