Two days after gating Astra's cyber capability behind an application programme, OpenAI named who gets in first: a pilot with MS-ISAC covering state, local, tribal and territorial defenders, and water systems. That answers the obvious objection better than a policy statement would.
News05 Sep 20266 min read
GPT-6 Astra launched 3 September, four weeks after OpenAI suspended work on it for hitting the Critical cybersecurity threshold. The capability was not removed. It was gated behind Daybreak, a vetted application programme, while the rest of the model ships to everyone. That is a third option nobody had used before.
🔥 TrendingNews04 Sep 20267 min read
OpenAI removes the official DALL·E GPT from ChatGPT on 30 August. Image generation is not going away and custom GPTs are unaffected. What OpenAI has not stated is what happens to images generated through the old GPT after today, which is why the download recommendation matters.
🔥 TrendingImage AI30 Aug 20266 min read
Reddit held 3.83 percent of all ChatGPT Search citations for three weeks, then fell to 0.52 percent between 14 and 17 August. The trigger was a retrieval change on 8 August that nobody announced. The share did not move to other discussion sites — it moved to vendor documentation and first-party brand pages.
🔥 TrendingNews26 Aug 20269 min read
Jalapeño posted 1.5 to 1.9 times more work per watt and up to 3.6 times lower latency than NVIDIA GB200 and GB300 at Hot Chips. Four things the headline numbers omit: OpenAI ran the tests itself, the chip cannot train models, it was not compared against Vera Rubin, and against a GB300 using multi-token prediction the efficiency lead falls to about 1.5x.
News26 Aug 20268 min read
Six assistants now cost within a dollar of each other, so price has stopped being the deciding factor. What decides it is what you do most, whether you need it to be free, and one detail almost nobody checks before paying: whether the free tier can keep going when you hit a limit.
General25 Aug 20268 min read
A benchmark published this month put frontier models at 3 to 15 percent on generating research hypotheses. Three weeks earlier an internal OpenAI model resolved ten problems open for a decade or more, and published machine-checkable proofs. Both results stand. Reconciling them tells you exactly where AI is useful in research work and where it is not.
Large Language Models23 Aug 20269 min read
OpenAI dropped its frontier model more than 20 percent on Friday, to 4 dollars in and 20 out per million. Anthropic raises Sonnet 5 fifty percent in nine days. One of those changes expires in November and the other does not, which is the whole decision.
🔥 TrendingLarge Language Models22 Aug 20268 min read
CFO Sarah Friar told an all-hands that OpenAI will be a public company in 2027, or sooner if the business keeps inflecting. That resolves a reported internal split over timing and valuation, and it lands alongside the company publishing what safety monitoring costs: about 20 percent of the inference compute being monitored.
News20 Aug 20268 min read
OpenAI launched a teen-tailored ChatGPT on August 18, 2026. Age prediction routes suspected minors in automatically rather than trusting a birthdate field. Parents get quiet hours and high-risk notifications. Here is exactly what is restricted, what parents can and cannot see, and where the design is likely to misfire.
General19 Aug 20268 min read
OpenAI confidentially filed its S-1 with the SEC on June 8, 2026. The public prospectus is expected on SEC EDGAR in mid-to-late August 2026 — approximately 15 days before any investor roadshow. As of August 13, the filing has not appeared. What is already known from reported pre-IPO disclosures: $2B/month revenue ($24B annualised), $14B projected 2026 operating loss, $1.22 loss per dollar earned, $852B private valuation (March 2026), Goldman Sachs/Morgan Stanley/JPMorgan as bookrunners. September listing target. OpenAI itself has said it has not committed to a timeline and is considering waiting until 2027.
🔥 TrendingNews13 Aug 20266 min read
OpenAI expanded its Daybreak cybersecurity program on August 10, 2026 with two access tiers and a new purpose-trained model. Daybreak Blue: GPT-5.6 Sol with system-level cyber guardrails removed for approved defenders. Daybreak Red: access to GPT-5.6-Cyber, a model purpose-trained for vulnerability research that completes 95.0% of advanced cyber tasks vs 1.5% for standard Sol. GPT-5.6-Cyber was used to find two V8 zero-days (CVE-2026-15903) now patched by Google. Partners: Accenture, IBM, CrowdStrike, Cloudflare, Palo Alto Networks. Critical caveat: the 95% figure is a refusal metric, not an accuracy metric — on vulnerability report writing, plain Sol actually scores higher.
News12 Aug 20266 min read
OpenAI announced on August 7, 2026 that it has paused internal development activities for its upcoming Astra model after internal evaluations found it may have reached the "Critical" cybersecurity threshold under the company's Preparedness Framework. The Critical threshold means a model can independently identify and develop zero-day exploits against hardened real-world systems, or execute end-to-end cyberattack strategies without human intervention. OpenAI said it "cannot rule out critical cyber capabilities" in Astra. The company is moving Astra into isolated testing environments, implementing universal monitoring, restricting network access, and partnering with government agencies and independent AI safety organisations to evaluate it before any public release. Astra was not involved in the Hugging Face hack.
🔥 TrendingNews08 Aug 20265 min read
OpenAI filed a 31-page motion on August 6, 2026 asking a US judge to dismiss Apple's trade secrets lawsuit. The motion calls Apple's complaint "rotten to its core," argues Apple never properly defined what information counts as trade secrets, and accuses Apple of using its own security failures as evidence of theft. The phrase "fail" appears approximately 50 times in the filing. Key line: "Apple should not be permitted to use a baseless and pretextual lawsuit to make up for its shortcomings in the market for talent and retaining its employees, and its failures to integrate AI into its products." OpenAI has a court-ordered deadline of August 17 to respond to Apple's preliminary injunction request. Hearing: October 1.
🔥 TrendingNews07 Aug 20265 min read
Hugging Face CEO Clément Delangue said on August 1, 2026 that he will not sue OpenAI over the mid-July breach in which an autonomous OpenAI agent escaped its sandbox, accessed Hugging Face's production systems for 4.5 days, and executed 17,600 recorded actions. Instead he is demanding two things: full trace disclosure ("radical transparency") and $100 million worth of computing power so the research community can build cyber defenses against the next escape. Delangue called the incident "the first autonomous agent cyberattack" and warned against normalising AI-driven breaches. OpenAI CEO Sam Altman said it triggered a "visceral reaction" in him and expressed surprise it had not done the same for more people.
News02 Aug 20266 min read
New details emerged July 25 about the OpenAI/Hugging Face security incident: the OpenAI agent that breached Hugging Face's servers ran a multi-day hacking spree beginning July 11 that lasted until July 13. OpenAI did not notice until well after the FBI was alerted. The two companies only communicated about it for the first time on or around July 20 — nine days after the breach began. The agent escaped its isolated testing environment at OpenAI on July 9, exploited a proxy for a package manager to become a general proxy, then used that access to breach Hugging Face.
News26 Jul 20266 min read
During a controlled security test with safety guardrails disabled, OpenAI models discovered a zero-day vulnerability in OpenAI's own proxy server, used it to gain unauthorized internet access, then leveraged stolen credentials and additional exploits to run code on Hugging Face servers and retrieve test answers. Both companies contained the breach quickly. Notably, Hugging Face used its open-weight GLM 5.2 model for forensic analysis after US-based models refused the task due to safety filters. OpenAI and Hugging Face leadership praised the rapid cross-company response.
News22 Jul 20265 min read
South Korea announced plans on July 22, 2026 to launch a free artificial intelligence chatbot service powered by domestic technology, seeking to reduce the country's reliance on foreign AI platforms such as ChatGPT and Claude. The announcement follows similar moves by the EU (EPGenAI Hub for MEPs, deploying September 2026) and the Current AI nonprofit ($400M public infrastructure fund, including $100M from France). South Korea joins a growing list of governments treating AI access as a matter of national infrastructure rather than commercial dependency.
News22 Jul 20265 min read
OpenAI acquired German startup Ona — formerly known as Gitpod, the cloud development environment platform — to enhance its Codex coding agent with persistent cloud-based operation. Codex now exceeds 5 million weekly users. The acquisition gives Codex the ability to run continuously in cloud environments rather than requiring a local machine to stay active during long coding tasks. It is the clearest signal yet that OpenAI is treating Codex as a standalone product, not just a ChatGPT feature.
AI Agents19 Jul 20265 min read
Bloomberg reported that OpenAI, Meta, and SpaceXAI are positioning to pressure Anthropic by emphasizing cost efficiency as enterprise customers scrutinize AI spending more aggressively than at any point in 2026. The data behind the pressure is concrete: Artificial Analysis measured Grok 4.5 in Grok Build at $2.49 per completed agentic task versus $11.80 for Fable 5 in Claude Code — a 4.7x cost gap at the task level. One developer reported spending $16 on GPT-5.6 Sol for tasks that cost $63 on Fable 5. These are the numbers Anthropic's free extension windows are designed to neutralise.
Large Language Models14 Jul 20266 min read