#openai Tech News.

OpenAI is spending $1 billion to subsidize access to its Daybreak security platform after unveiling Astra, an AI that can independently find previously unknown software flaws and turn them into working attacks.

OpenAI launched GPT-6 Astra, a new frontier AI model for computer use, professional work, science, coding, and cybersecurity. The model can browse the web, build websites, and handle complex tasks with less human intervention, while presenting significant cybersecurity capabilities and risks.

OpenAI pledged $1 billion in subsidized access, training, and technical support for cyber defenders through its Daybreak for Frontline Defenders initiative. The program initially targets U.S. operators of essential services, including water utilities, electric grid operators, and community banks, with plans to expand to partner countries.

Cloudflare opened early access to Vulnerability Discovery and Remediation, using OpenAI's GPT-5.6-Cyber model to find and block code flaws. The service runs in Cloudflare Managed Defense, requires human approval for fixes, and addresses a record 60,475 vulnerabilities logged by September.

OpenAI released GPT-6 Astra, its most intelligent model, initially to Trusted Access Program members, with Plus, Pro, Business, and Enterprise subscribers gaining access in coming days. It excels at computer-use, browsing, software engineering, and cybersecurity, scoring 98.6% on ARC-AGI-3 and a perfect ExploitBench score.

OpenAI launched GPT-6 Astra, its most capable model, with gains in software engineering, computer use, and cybersecurity. It is the first to hit the "Critical" cybersecurity threshold and is more jailbreak-resistant than GPT-5.6 Sol. President Greg Brockman personally thinks "we're there" on AGI.

Sam Altman says the AI industry has done a terrible job communicating the technology's benefits, leaving developers exposed to backlash. Local opposition blocked or delayed 75 data centre projects worth about $130B in one quarter. Denmark's Energinet paused grid connections, holding 60GW of queued projects against 7GW peak demand.

Anthropic, OpenAI, and xAI chatbots suffered rare outages Thursday morning. SpaceX attributed Grok's issues to an outage at its Memphis compute center. OpenAI cited a routing error starting at 7:43 am PT, resolved by 8:17 am PT. Anthropic resolved its partial outage by 9:16 am PT.

OpenAI released GPT-6 Astra, its most intelligent model, excelling in cybersecurity and software engineering. Anthropic updated Fable 5.1 and Mythos 5.1, Meta launched Muse Spark 1.3, and Google introduced Gemini 3.8 Flash and 3.8 Flash Cyber, all focusing on agentic workflows.

GPT-6 Astra, OpenAI's first Stargate and lightly looped supermodel, launched today, beating Fable 5.1 on FrontierMath (97.6%) and ARC-AGI-3 (99.9%). Early testing burned over 20B tokens, confirming Astra can automate AI engineering tasks at 33 tokens per second for under $6 an hour.

Cloudflare announced early access to Vulnerability Discovery and Remediation, part of Cloudflare Managed Defense. The invitation-only service uses OpenAI Daybreak models, including GPT-5.6 Cyber, through the OpenAI Daybreak Defense Network to detect, prioritize, and propose fixes for code vulnerabilities, with no changes taking effect without human approval.

SentinelOne expanded its Wayfinder Frontier AI services by integrating OpenAI Daybreak models. The company announcement details the addition of these models to the platform. This move broadens SentinelOne's AI-powered security offerings for its customers.

OpenAI commits $1 billion in subsidized model access, training, technical support, and partnerships to a new initiative protecting essential services worldwide. The funding expands subsidized access to models for participating organizations.

OpenAI is launching an initiative providing subsidized model access to water systems, electricity providers, local governments, and other critical services. President Greg Brockman announced the program at a summit hosting 300 security leaders at company headquarters Thursday, addressing anticipated AI-enabled cyberattacks.

Sen. Bernie Sanders announced legislation for a permanent ban on superintelligence and a pause on advanced AI development. The bill, co-announced with Rep. Greg Casar, would create a federal regulator and impose up to 20 years in prison for violators. OpenAI released GPT-6 Astra the same day.

OpenAI's PrimeGaps186 repository formalizes in Lean 4 that every admissible 40-integer tuple has infinitely many translates with at least two primes, yielding a gap bound of 186. The proof is conditional on three explicit axioms and a Python numerical certificate.

GPT-6 Astra scored 98.6% on the ARC-AGI-3 benchmark, up from GPT-5.6 Sol's 7.8%. Humans navigated the interactive settings in March while frontier models scored under 1%. The benchmark tests uncharted interactive territory, not training-data recall.

OpenAI released GPT-6 Astra on Thursday, built on its largest-ever training run using more than 100,000 GPUs at the Stargate site in Texas. President Greg Brockman called it a "generational leap" that could eventually be seen as the arrival of AGI.

OpenAI priced GPT-6 Astra at $10 per million input tokens and $50 per million output tokens, matching Anthropic's pricing for Claude Fable 5.1. The model, part of the Hugging Face breach, will be released with new safeguards.

The Trump administration filed a brief in Manhattan federal court supporting OpenAI's fair use defense against The New York Times' copyright lawsuit. The Justice Department argued AI training on copyrighted works outweighs competitive harm, while the Times accused the administration of siding with trillion-dollar AI companies over creators.

OpenAI, Anthropic, Google, Microsoft, Amazon, Oracle, and over 100 other companies signed a letter urging a global surge in cyber defense, warning that AI-enabled attacks will become more widespread and sophisticated in coming months, putting hospitals, water plants, and internet infrastructure at risk.