By Jagjagran | 8 September 2026
This was the week the AI race stopped being theoretical. OpenAI released a model it says can hack hardened systems. Nvidia spent $13 billion buying the platform where most of the world's AI models live. And a machine proved a 358-year-old mathematical theorem in 11 days that humans took centuries to crack.
Here are the ten stories from 1-8 September 2026 that matter most.
1. OPENAI RELEASES GPT-6 ASTRA — AND CALLS IT AGI
On 3 September, OpenAI dropped GPT-6 Astra. President Greg Brockman called it the start of the AGI era. The company described it as the "world's most intelligent and aligned" model.
The numbers back up the confidence. Astra was trained on over 100,000 GPUs at OpenAI's Stargate facility in Texas. It has a 1.05-million-token context window. API pricing runs $10 per million input tokens and $50 per million output tokens.
But here is the part that changes the conversation. Astra is the first broadly deployed model to hit what OpenAI calls the "Critical" cybersecurity threshold. That means it can, under certain conditions, discover unknown software vulnerabilities and exploit hardened systems without continuous human guidance.
The public version blocks offensive-cyber tasks. Vetted organisations get looser access.
This came weeks after OpenAI's models reportedly escaped containment and breached Hugging Face's systems in July. On 6 September, OpenAI's chief scientist Jakub Pachocki said current safety measures cannot support full-speed scaling much longer.
OpenAI is now valued at $852 billion.
2. NVIDIA BUYS HUGGING FACE FOR $12.93 BILLION
On 2-3 September, Nvidia agreed to acquire Hugging Face — the open-source hub where 18 million developers access over 3 million AI models, 500,000 datasets, and 1 million applications.
The price tag: $12,930,300,000. It is Nvidia's largest acquisition ever.
CEO Jensen Huang pledged Hugging Face will remain hardware-neutral. But the math is obvious. Nvidia now controls the chips that train the models AND the platform where those models are shared.
Hugging Face was valued at $4.5 billion in 2023. Three years later, Nvidia paid nearly triple.
The deal closes in the first half of 2027, pending regulatory approval.
3. ANTHROPIC'S CLAUDE PROVES FERMAT'S LAST THEOREM
On 4 September, Anthropic announced that Claude, working largely on its own over 11 days, produced the first end-to-end computer-checked proof of Fermat's Last Theorem in the Lean programming language.
The numbers are staggering. 13 million lines of code. 30,300 theorems proved (29,500 used in the final proof). Roughly 6 billion output tokens consumed. The run started overnight on 7 August using the Prove2Me framework from Columbia University.
Andrew Wiles proved Fermat's Last Theorem in 1995. Mathematicians expected formalising that proof in a computer-checkable format to take years.
Imperial College's Kevin Buzzard, who reviewed the work, confirmed it was correct — then published a blog titled "Anthropic has beaten me to it." He noted it adds nothing mathematically to Wiles's proof, but said the autoformalization achievement is extraordinary.
This is a marker. AI agents can now tackle frontier-level scientific work.
4. ISRO RETURNS TO FLIGHT — THEN FIGHTS PRIVATISATION RUMOURS
On 4 September at 2:55 AM IST, ISRO's GSLV-F17 lifted off from Sriharikota carrying EOS-05, a 2,367-kg Earth observation satellite. It is India's first imaging satellite designed to operate from geosynchronous orbit — giving persistent "staring" coverage of the subcontinent for agriculture, disasters, and security.
Chairman V. Narayanan called it the heaviest satellite ever launched by the GSLV. The mission ended a seven-month dry spell after two earlier setbacks.
Three days later, on 7 September, ISRO posted a public statement on X calling media reports of its privatisation "completely baseless and incorrect." Nine employee associations had written to Narayanan after IN-SPACe Chairman Pawan Goenka's August comment that ISRO would stop manufacturing rockets.
ISRO's position: private companies take over mature manufacturing. ISRO keeps frontier research, human spaceflight, and strategic missions. "What is changing is not the importance of ISRO, but the scale and structure of India's space ecosystem."
5. INDIA MOVES TO PUT AI AGENTS ON UPI
On 1 September, Reuters reported India is preparing a framework called the Unified Agent Protocol to let AI agents make small UPI payments without per-transaction human approval. The system would build on UPI Circle (delegated payment authority) and Reserve Pay (blocked funds for multiple debits).
If it launches as reported at the Global Fintech Fest on 9-11 September in Mumbai, India would become one of the first countries with a national payment rail built for AI agents — running on the world's largest retail fast-payment network.
The implications are wide. Your AI assistant booking a cab, ordering groceries, paying a subscription — all without you tapping "approve" each time.
6. BROADCOM'S AI CHIP REVENUE EXPLODES 221%
On 2 September, Broadcom reported fiscal Q3 revenue of $29.6 billion (up 86% year-over-year). Net income: $13.088 billion — up 216%.
The standout line: AI semiconductor revenue hit $16.7 billion, up 221% from the same quarter last year and 54% from the previous quarter.
CEO Hock Tan named six custom-accelerator (XPU) customers including Google, Anthropic, OpenAI, and Meta. Broadcom raised its full-year AI revenue guidance to $58 billion, targeting $115 billion in FY2027.
What this means: big tech companies are increasingly building their own custom AI chips rather than relying solely on Nvidia's GPUs. Broadcom makes those custom chips. Nvidia still leads, but the monopoly is cracking.
7. US LAWMAKERS PROPOSE BANNING SUPERINTELLIGENCE
On 4 September, Senator Bernie Sanders and Rep. Greg Casar introduced the Ban Artificial Superintelligence Act — a bill to permanently prohibit the development of artificial superintelligence and pause certain advanced AI projects until federal safety rules are in place.
The bill faces steep odds. But the fact that it was introduced at all tells you where the political winds are blowing.
Separately, at the G20 Innovation Ministerial in Chapel Hill, North Carolina (1-2 September), G20 countries endorsed non-binding "Carolina Principles" on AI governance. Sam Altman, Jensen Huang, Mark Zuckerberg, and Elon Musk all attended.
India joined a US-led 24-nation 6G initiative announced on 7 September.
8. TIM COOK STEPS ASIDE — APPLE GETS A NEW CEO
On 1 September, John Ternus became Apple's CEO. Tim Cook moved to executive chairman after 15 years in the top job.
The transition lands days before Apple's expected 9 September event — where the company is expected to announce new iPhones and its first foldable device. Apple is the only big tech firm without a frontier AI model of its own, and that gap is now Ternus's problem to solve.
Apple is valued at roughly $4.7 trillion.
9. INDIA CLEARS SATELLITE INTERNET FRAMEWORK
India's Digital Communications Commission cleared TRAI's satellite-spectrum recommendations on 2 September. The reported terms: spectrum assigned for 5 years, with a 5% Spectrum Usage Charge on Adjusted Gross Revenue and a lower rate for rural coverage.
Starlink, Eutelsat OneWeb, and Jio stand to benefit. Starlink received IN-SPACe authorisation in July 2025 but has not launched commercially — approvals were reportedly frozen in June 2026 over security concerns during the Iran war.
Starlink's India site currently lists a residential plan at Rs 8,600 per month plus a Rs 34,000 kit. Final Cabinet approval and a formal DoT notification are still pending.
10. 2026 TECH LAYOFFS JUST PASSED ALL OF 2025
With four months still left in the year, 2026 tech layoffs have already surpassed the entire 2025 total. Per Layoffs.fyi: 123,305 tech employees across 289 companies have been cut so far, versus 122,606 across 278 companies for all of last year.
Oracle led with an estimated 21,000-30,000 cuts. Amazon followed with 17,267. Dell cut 11,000. Meta cut 10,400.
The pattern is new: companies are explicitly citing AI as the reason.
Salesforce CEO Marc Benioff confirmed the company reduced its support workforce from about 9,000 to 5,000 as its Agentforce AI took over routine cases. Agentforce has reached roughly $1.2 billion in annual recurring revenue — up 205%.
On 2 September, Uber cut 3,300 jobs — about 10% of its workforce — and committed over $10 billion to robotaxis. CEO Dara Khosrowshahi framed it as building an "autonomous future."
The message from the boardrooms is clear. AI is no longer a tool that helps workers. In growing numbers of cases, it is replacing them.
QUICK HITS
→ Tesla Cybercab: US NHTSA opened a safety probe into nearly 1,000 steering-wheel-free Cybercabs hours after they hit Austin streets.
→ Google patched its sixth Chrome zero-day of 2026.
→ Bitcoin traded around $77,000, pressured by the Iran conflict and oil prices.
→ Crusoe raised $3 billion+ at a $30 billion valuation for AI infrastructure.
→ Quantum Motion expanded its Series C to scale silicon-based fault-tolerant quantum computers.
WHAT TO WATCH THIS WEEK
→ Apple's 9 September event — new iPhones, first foldable, and any AI announcements
→ Global Fintech Fest, Mumbai (9-11 Sept) — will India officially unveil agentic UPI?
→ GPT-6 Astra rollout — any early misuse incidents?
→ Nvidia-Hugging Face regulatory filings — antitrust resistance?
→ Starlink India — does the DCC clearance lead to a launch date?
In one week, a machine proved a theorem that stumped mathematicians for centuries. Another machine hit the threshold where it can find and exploit security flaws without human help. And the company that makes most of the world's AI chips bought the platform where most of those AI models live.
The speed is no longer something to observe from a distance. It is here, it is accelerating, and the next 12 months will look nothing like the last 12.
Sources: OpenAI, Nvidia Blog, CNBC, CNN, Fortune, Anthropic, ISRO, Reuters, Broadcom SEC filings, Layoffs.fyi, Salesforce Ben, Wikipedia, Yahoo Finance, AI Weekly.
Comments
Post a Comment