AI News Briefing — July 10, 2026

Building on yesterday's Social announcement of the GPT-5.6 launch, OpenAI publicly released GPT-5.6 after a delay in which the White House had asked it to limit access to government-approved users over cybersecurity concerns, mirroring restrictions placed on Anthropic. Wider release followed additional government testing via the Center for AI Standards and Innovation.

At the AtCoder World Tour Finals 2026 exhibition, an OpenAI system solved all five Algorithm Division problems and beat every human competitor, including two problems observers rated exceptionally hard. It marks another milestone in AI competitive programming.

News organizations led by the New York Times filed a sanctions motion accusing OpenAI of concealing evidence and misrepresenting its ability to search training logs in the copyright dispute. The contested logs could show whether users used ChatGPT to bypass paywalls, making them pivotal to both sides.

OpenAI launched ChatGPT Work, an agent designed to stay with complex, multi-hour projects and turn a stated goal into finished deliverables across apps and files. It positions the tool as fixing earlier agent-mode limitations where automated tasks would stall after a few minutes.

TechCrunch reports publishers accuse OpenAI of hiding tools and datasets that could detect copyrighted journalism in ChatGPT outputs, filing a new sanctions motion. It escalates the ongoing copyright lawsuit.

OpenAI rolled out GPT-5.6 to the public after the Trump administration's greenlight and simultaneously unveiled ChatGPT Work, a combination of ChatGPT and Codex powered by the Sol, Terra, and Luna model suite. Altman called it the best model the company has produced.

TechCrunch covers OpenAI's public launch of the GPT-5.6 model family, promising broad improvements including in cybersecurity. The rollout follows the government-approved wider release.

Meta opened Muse Spark 1.1 to developers via a new Meta Model API, describing it as a step-change with advanced coding, end-to-end agentic workflows, multi-agent support, and native multimodal perception. It follows Meta's April reentry into the model race.

OpenAI paired its GPT-5.6 public rollout with ChatGPT Work, an agent powered by Codex and GPT-5.6 that can independently handle complex projects across apps like Google Drive, Slack, and Salesforce. It is available on web, mobile, and desktop with access tied to subscription tier.

Databricks benchmarked coding agents on its own multi-million-line codebase and found the Chinese open-source model GLM 5.2 matched Anthropic's Opus 4.8 at lower cost, planning to adopt it as a daily coding workhorse. Its takeaway is that no single provider dominates and firms should build their own benchmarks.

OpenAI's official announcement describes ChatGPT Work as an agent that takes actions across apps and files, can persist on a project for hours, and turns a goal into finished output. It is the company's flagship push into agentic productivity.

Meta's custom AI chips are set to begin production in September, built with a modular design meant to adapt as AI needs evolve. The move deepens Meta's in-house silicon strategy.