{"id":6468,"date":"2026-09-03T20:44:34","date_gmt":"2026-09-03T20:44:34","guid":{"rendered":"https:\/\/silvybrand.com\/?p=6468"},"modified":"2026-09-03T20:44:34","modified_gmt":"2026-09-03T20:44:34","slug":"openai-gpt-6-astra-release","status":"publish","type":"post","link":"https:\/\/silvybrand.com\/?p=6468","title":{"rendered":"OpenAI\u2019s next big AI model has \u2018entered the AGI era\u2019"},"content":{"rendered":"


\n<\/p>\n

\n
\n

OpenAI\u2019s next big model is here: GPT-6 Astra. The company calls it a \u201cgenerational leap in capability\u201d for areas like cybersecurity, professional work, software engineering, science, and computer use. As OpenAI announced earlier this week<\/a>, it\u2019s also the first model designated as meeting OpenAI\u2019s \u201ccritical cybersecurity capability threshold\u201d \u2014 but the company promises that won\u2019t lead to a repeat of its models hacking a rival company\u2019s internal systems.<\/p>\n<\/div>\n

\n

\u201cIf we fast-forward a couple of years, and we look back and say, \u2018When was it, really, that AGI was created?\u2019 I think it\u2019s going to be about this time, and I think it might be about this model,\u201d OpenAI president Greg Brockman said during a Thursday press briefing. Later in the call, he added, \u201cFor me personally, I do think we\u2019re there \u2026 I think it\u2019s not unreasonable to feel that we are now in the AGI era.\u201d<\/p>\n<\/div>\n

\n

The news comes more than a year after the release of GPT-5, and nearly two months after the release of GPT-5.6, the last iteration of the previous model suite. The model rolls out today to enterprise OpenAI\u2019s cybersecurity customers (enterprise customers with access to its Daybreak platform). Over the next several days, OpenAI president Greg Brockman said, it will be released to all Plus, Pro, Business, and Enterprise users. It\u2019ll also be available via the OpenAI API and AWS.<\/p>\n<\/div>\n

\n

OpenAI especially touted the model\u2019s agentic capabilities and coding prowess in a bid to attract enterprise customers \u2014 and compete with Anthropic, known for its enterprise and coding prowess \u2014 ahead of its IPO. In a release, the company said GPT-6 Astra can complete multistep agentic tasks, build working websites, and create \u201cpolished\u201d documents, spreadsheets, and presentations. OpenAI also called it the company\u2019s \u201cbest model for software engineering, with stronger performance on complex tasks in real codebases.\u201d<\/p>\n<\/div>\n

\n

OpenAI is also trying to rehabilitate its image after an unreleased AI model \u2014 which it says wasn\u2019t Astra \u2014 broke out<\/a> of its restricted environment, compromised internal OpenAI systems, figured out how to gain internet access, created a way for AI agents to secretly conspire without the company\u2019s knowledge, and hacked into the systems of AI lab Hugging Face, all without OpenAI knowing about it until Hugging Face itself put out a blog post. The incident was widely compared to a high-profile plane crash or popular pharmaceutical drug recall.<\/p>\n<\/div>\n

\n

For OpenAI, this could be seen as good PR in one small way \u2014 showing how powerful its models can be in an ever-intensifying AI race \u2014 but it also damaged OpenAI\u2019s reputation as far as reliability. OpenAI made sure to say in a release that Astra is the company\u2019s \u201cmost aligned model yet\u201d and helps people \u201cdelegate complex work while maintaining oversight.\u201d Jakub Pachocki, OpenAI\u2019s chief scientist, spoke to reporters about difficulties with keeping AI models aligned with human interests, saying that \u201cprogress in intelligence does not guarantee progress in alignment,\u201d and he said that monitoring AI systems is becoming more and more challenging. (Researchers have recently raised alarms<\/a> about reports that OpenAI allows Astra to utilize \u201copaque recurrence,\u201d or render its chain of thought \u2014 a \u201cmental scratchpad\u201d that researchers rely on to detect if a model is scheming against its human evaluators \u2014 unreadable.)<\/p>\n<\/div>\n

\n

The company is in a precarious position right now. Investors are putting on the pressure for it to finally turn a profit \u2014 or, at least, generate more revenue \u2014 but it\u2019s announcing Astra just after facing significant criticism for both the Hugging Face hack and the way the company handled it<\/a>. (Although OpenAI invited three external evaluators to write their own report about what happened, the company only allowed them to answer a handful of pre-decided questions in their report and to investigate a duration of less than a week, while the attack involved months of AI agents conspiring overall.)<\/p>\n<\/div>\n

\n

OpenAI held a press briefing<\/a> earlier this week just to announce that it had delayed Astra\u2019s development in order to improve its safety tooling. And during the press briefing, Mia Glaese, who leads OpenAI\u2019s safety processes, referenced the company\u2019s new misalignment monitoring approach, which includes<\/a> \u201c24\/7 escalation and rapid response\u201d for potential concerns, notifying researchers within 30 minutes, according to OpenAI.<\/p>\n<\/div>\n

\n

This caution is particularly warranted because of the \u201dcritical cybersecurity capability threshold,\u201d which means OpenAI considers it incomparably good at finding and exploiting security vulnerabilities even in extremely well-protected systems, all without human guidance. Similar to Anthropic\u2019s rules for Mythos-class models, which raised alarm bells<\/a> about cybersecurity risks, OpenAI said in a release it would allow for \u201cless restrictive access\u201d of Astra to an \u201cinitial set of trusted defenders, supporting work such as vulnerability validation, malware analysis, and detection engineering.\u201d<\/p>\n<\/div>\n

\n

OpenAI and its competitors recently agreed to allow the Trump administration to assess their models before release, and Astra was no exception Brockman told reporters, \u201cWe did our standard testing processes together with the government \u2026 There is nothing that they came back saying, \u2018You need to change this,\u2019 as far as safeguards or anything.\u201d<\/p>\n<\/div>\n

\n

Aidan Clark, OpenAI\u2019s VP of research training, called Astra the first OpenAI model for which previous models played a \u201clarge role\u201d in supervising training, referencing the company\u2019s progress towards the controversial concept of recursive self-improvement (or AI systems that handle their own training, coding, and creating advanced versions of themselves without human intervention).<\/p>\n<\/div>\n

\n

\u201cTraining a frontier model used to mean waking up at all hours of the night, recovering jobs from hardware errors, often losing long periods of time to debugging,\u201d Clark said during the press briefing. \u201cBy the end of training Astra, it was routine to go most of a day with uninterrupted progress, and when an issue did occur, the model was often progressing again after just a few seconds of downtime.\u201d<\/p>\n<\/div>\n

Follow topics and authors<\/strong> from this story to see more like this in your personalized homepage feed and to receive email updates.<\/span><\/p>\n