OpenAI and Anthropic public-post watch

Checked 10 August 2026 at approximately 3:02 am AEST (5:02 pm UTC on 9 August). Priority window: the preceding eight hours.

Eight-hour headlines

• Codex understands an informal /loop instruction even though no such command exists — OpenAI’s Jason Liu said that sending the literal string /loop is enough for Codex to infer the intended behavior. This is an undocumented interaction pattern, not a newly shipped slash command, but it is a useful developer signal: natural-language intent can substitute for a formal command when users want the agent to continue iterating. Original post

• OpenAI is leaning on practitioner education to drive ChatGPT Work adoption — Greg Brockman pointed users to a 61-minute walkthrough that presents Work as “Codex in the cloud” across phone, web, and desktop, covering 14 knowledge-work capabilities including presentations, plugins, and Blocks. Brockman’s post does not announce a feature, but the endorsement shows the company promoting a broad, agentic workflow rather than a single-purpose chat experience. Original post

Other important posts since the previous run

• ChatGPT Finance surfaced a concrete subscription-auditing use case — Brockman amplified a user’s report that the product found $550 per year in forgotten or supposedly cancelled subscriptions. The post landed about 50 minutes before the priority window, so it is included as a near-window product signal rather than a fresh launch or independently verified performance claim. Original post

• An OpenAI product lead reset usage limits for all paid ChatGPT Work and Codex users — Tibo Sottiaux announced the one-off reset while emphasizing that GPT-5.6 Sol can run in other agent harnesses, including Claude Code. In the same exchange, Anthropic’s Boris Cherny said Anthropic does not ban people merely for using its harness with other models; he attributed a reported suspension to a likely unrelated account classifier and said the company was investigating. Together, the posts are a useful interoperability and account-enforcement signal, though neither establishes a general guarantee about every third-party configuration. OpenAI post · Anthropic reply

• GPT-5.6 Luna’s low price is enabling very large exploratory runs — Romain Huet highlighted a user’s report of processing almost one billion tokens for roughly $30. This is an anecdotal usage example rather than an OpenAI benchmark, but it illustrates the workloads developers are attempting after Luna’s price reduction. Original post

Coverage note

Public X timelines were checked for Sam Altman, Greg Brockman, Romain Huet, Nick Turley, Kevin Weil, Mark Chen, Noam Brown, Sebastien Bubeck, Boaz Barak, Brian Huang, Jason Liu, Will Depue, Steven Heidel, Philipp Spiess, Michelle Pokrass, Tibo Sottiaux, Atty Eleti, Dario Amodei, Jack Clark, Amanda Askell, Alex Albert, Boris Cherny, Jan Leike, Chris Olah, Mike Krieger, Trenton Bricken, and Eric Jang. No substantive Anthropic employee post was visible inside the strict eight-hour window; Cherny’s clarification above was posted earlier but after the previous bulletin check. Previously covered Astra, Claude Code auto-mode, GPT-Live, inter-agent communication, and safety-regulation items were not repeated.