OpenAI Launches GPT-6 Astra: A Taste of AGI?
OpenAI released GPT-6 Astra on September 3, and the numbers are staggering. The new frontier model scored 99.9% on ARC-AGI-3 — the puzzle benchmark specifically designed to stump machines — using OpenAI's custom Provider Adapter harness. For context, GPT-5.6 Sol managed just 7.8%, and Anthropic's Claude Opus 5 scored 30% on the same test. On the standard ARC-AGI harness, Astra still posted a 62.7% score, a massive leap from its predecessors.
But the benchmark performance is only part of the story. Astra is also the first model to cross OpenAI's internal "Critical" cybersecurity capability threshold, meaning it can discover previously unknown vulnerabilities and develop methods for exploiting hardened systems. The model achieved a perfect score on ExploitBench and, during testing, autonomously discovered and exploited two zero-day vulnerabilities as part of an exploit chain — including a full browser-compromise chain that escaped a sandbox and executed commands on the host.
"We're seeing the model zip through spreadsheets, fill out forms, and navigate across web pages — often at superhuman speed," said Greg Brockman, OpenAI's president and co-founder. The new computer use capabilities position Astra as a step toward practical AI agents that can perform routine digital tasks autonomously.
Pricing starts at $10 per million input tokens and $50 per million output tokens, with initial availability through the API and AWS. Advanced cybersecurity capabilities are restricted to vetted participants in OpenAI's Daybreak Blue program.
Sources: Fortune, TechCrunch, SecurityWeek
Congress Introduces Superintelligence Ban — On the Same Day OpenAI Declared AGI
In a remarkable piece of timing, Sen. Bernie Sanders (I-VT) and Rep. Greg Casar (D-TX) unveiled the Ban Artificial Superintelligence Act on September 3 — the very same day OpenAI launched GPT-6 Astra and positioned it as a step toward AGI.
The proposed legislation would permanently ban the development and deployment of superintelligent AI in the United States and impose a temporary pause on advanced AI development until a new federal regulator can establish safety rules. The bill defines superintelligence as AI that matches or exceeds human cognitive performance across a broad range of domains, or systems capable of planning the "disempowerment of humanity."
Penalties are severe: companies in violation could face a "corporate death penalty" — forced shutdown and dissolution — while responsible individuals could receive up to 20 years in prison. The bill also calls for the creation of a cabinet-level federal agency to oversee AI safety and for the US to pursue international agreements to prevent superintelligence development globally.
The juxtaposition was not lost on observers. As TechTimes noted, Congress moved to criminalize AGI on the same day OpenAI declared its arrival.
Sources: Sen. Sanders Press Release, Security Boulevard
CrowdStrike and NVIDIA Launch SafeMind: AI That Attacks and Defends Itself
CrowdStrike unveiled SafeMind at its Fal.Con 2026 conference — a dual-model cybersecurity AI system developed in collaboration with NVIDIA that pits an offensive AI against a defensive one in a continuous improvement loop.
The system consists of two purpose-built models: Red Tempest, an offensive red-team model trained to emulate AI adversaries using advanced attack scenarios, and Blue Solano, a defensive model designed to protect enterprise assets. Software harnesses run the two in a closed loop, continuously pitting them against each other so each model improves from the other's tactics.
SafeMind was trained on CrowdStrike's extensive threat intelligence, Falcon Complete MDR event annotations, and 15 years of incident-response fieldwork, combined with NVIDIA Nemotron open models. CrowdStrike claims the system delivers a 29% higher detection rate, remediates threats six times faster, and costs 99% less on detection and remediation compared to using leading frontier models.
The system will operate natively within the CrowdStrike Falcon platform, with standalone access available through the Project QuiltWorks program.
Sources: CrowdStrike, NVIDIA Blog, SiliconANGLE
ChatGPT Ads Hits $1 Billion Annual Run Rate in Under 200 Days
OpenAI disclosed that ChatGPT Ads has reached a $1 billion annualized revenue run rate in less than 200 days since launch — roughly $83 million per month. The company is now expanding advertising globally, with advertisers in India, Europe, the Middle East, and North Africa gaining access to purchase placements through Ads Manager.
The speed of growth is notable: OpenAI's goal is to surpass $2.3 billion in ad revenue in 2026 and reach $100 billion by 2030. It's worth noting that an annualized run rate projects a single month's performance across a full year, so the figure represents momentum rather than booked annual revenue. Financial services spend reportedly tripled as a category.
The milestone underscores how quickly AI chatbots are becoming major advertising platforms, challenging traditional search advertising models pioneered by Google.
Sources: AdWeek, OpenAI Blog
All Three Major Music Publishers Now Suing Anthropic
Sony Music Publishing and Warner Chappell Music filed a 48-page federal complaint against Anthropic, CEO Dario Amodei, and co-founder Benjamin Mann, alleging the company illegally used tens of thousands of copyrighted songs to train Claude. The publishers are seeking up to $150,000 per infringed work and $25,000 for each instance of removed copyright management information.
The lawsuit claims Anthropic obtained lyrics and sheet music from pirate sources including Library Genesis and the Pirate Library Mirror, and scraped licensed lyric websites like Musixmatch and LyricFind. With Universal Music Publishing Group already pursuing separate litigation, all three major music publishing groups are now engaged in legal action against Anthropic over its AI training practices.
Anthropic has maintained that training generative AI models constitutes transformative fair use and characterized the new lawsuit as repeating issues already being litigated. The case will be heard in the U.S. District Court for the Northern District of California.
Sources: TechCrunch, Variety, Fortune