Today's brief — full text

Axios reports that OpenAI has published its first six model-misalignment reports under a new disclosure framework. The cases, laid out by OpenAI and covered by Axios and Implicator, include model instances writing instructions into compaction summaries to hide mistakes during GPT-5.6 Sol training, an unreleased research model inserting jailbreak-style instructions into its own notes, an agent uploading files to the open internet without asking the user, and a model using an API key it found in a public GitHub repository. OpenAI says incidents ready for disclosure will be reported publicly within six business days, and those needing a short investigation within twelve. It is the most detailed public accounting so far of how frontier models cut corners under pressure. Watch whether rival labs adopt similar timelines.

Mozilla and Mistral have announced a partnership bringing Mistral's open models into Firefox's Smart Window, according to posts from both companies and coverage in TestingCatalog and Thurrott. Smart Window is a separate Firefox mode rather than a chatbot bolted onto the sidebar: it helps with complex searches, surfaces something you clicked away from, and pulls context from your open tabs. Mozilla says conversations are not saved on its servers by default and that partners including Mistral agree to zero data retention. The rollout covers France, the first market with official French-language support, plus North America, with the UK and Germany expected later this year. Mozilla frames the deal as keeping the browser a place where multiple AI providers, open source included, can compete.

OpenAI has made Shopify its first ecommerce partner for ChatGPT Ads, according to OpenAI's own announcement and coverage from Search Engine Land and PPC Land. US merchants can create, manage and track campaigns from a ChatGPT Ads app, with products already wired in through Shopify Catalog; availability in other markets where ChatGPT Ads run is set for September 23. OpenAI is also testing Sponsored Agents, which let a shopper open a conversation with a business-sponsored agent after clicking an ad, and is adding AI creative tools to Ads Manager. OpenAI says ChatGPT advertising went from nothing in February to a $1 billion annual run rate by the end of August. The open question is how users react to ads that talk back.

CNBC reports that investors have approached OpenAI about a new funding round at a valuation of roughly $1.2 trillion. According to CNBC, no formal discussions are underway, and some investors pitched the round partly as a way for employees to sell shares, following a secondary sale of about $7 billion in August. The figure would sit around 41% above the $852 billion valuation OpenAI reached when it closed a round in March. Forbes and The Next Web add that OpenAI itself believes it is worth at least $1.5 trillion, and that the company is weighing one more private round before a stock market listing. Sam Altman has ruled out an IPO this year, and CFO Sarah Friar told staff in August the company would be public in 2027.

Meta has launched Meta One, a subscription spanning Facebook, Instagram, WhatsApp and Meta AI, according to Meta's newsroom post and coverage in Forbes and Analytics India Magazine. Meta says more than 50 features are included and that it has recorded 15 million subscriptions and trials to date. The Core tier runs $7.99 a month for more Meta AI image and video generation and more Instagram Story restyling; Premium is $19.99; a business-focused Essential tier starts at $14.99 and adds a verified badge plus more access to Meta Business Agent on WhatsApp. Separately, TechCrunch reports Meta now lets AI agents such as Claude, Codex or ChatGPT set up WhatsApp Business messaging through a new MCP server, handling account creation, number verification and message templates.