Llm

• Accel is in talks to lead a $1 billion funding round for Thinking Machines Lab, a startup founded by former OpenAI chief Mira Murati. • The deal would value the company at $40 billion, supported by an annual revenue run rate exceeding $100 million. • This investment would establish Thinking Machines as one of the world's most valuable AI startups and validate the commercial viability of frontier model development.

• OpenAI launched its GPT-6 "Astra" model on September 6, 2026, but users jailbroke the system within 24 hours. • Accel is reportedly in talks to lead a $1 billion funding round for Mira Murati's Thinking Machines Lab at a $40 billion valuation. • Thinking Machines Lab has reached an annual revenue run rate exceeding $100 million, marking a high growth benchmark for an early-stage startup.

• OpenAI has launched GPT-6 Astra, its largest frontier model to date, featuring state-of-the-art capabilities for coding and computer use. • The new model is 2.5x more expensive per token, though it is reported to be significantly cheaper per completed task. • Professionals may see higher efficiency in complex workflows due to reduced costs per task, despite the model being less monitorable.

• OpenAI is preparing to launch "Astra," a new model utilizing a "recurrent depth" technique that allows it to reason beyond standard sequential thinking patterns. • AI safety experts have raised alarms, warning that this novel architecture may be significantly harder to audit and constrain than previous models. • The company also previewed cybersecurity precautions after Astra demonstrated a strong capability for breaking into computer systems.

• AI model-training startup AfterQuery has become Y Combinator's fastest-ever unicorn, reaching a valuation of $3.2 billion. • The company's valuation saw a massive 10x increase in just five months, jumping from $300 million following its $30 million Series A round in April. • Separately, a specialized MiniMax H3 LoRA leaderboard has launched to provide a benchmarking resource for fine-tuned variants of the MiniMax H3 architecture.

• The U.S. government has filed a 20-page brief supporting OpenAI in its legal battle against The New York Times over the unlicensed use of copyrighted material to train large language models. • The administration argues that maintaining a robust and competitive domestic AI industry is a critical national interest to ensure the U.S. sets global standards for AI practice. • This intervention signals a significant shift toward favoring AI developers over copyright holders, potentially lowering the legal barriers for training generative AI on existing data.

• Security researchers discovered a vulnerability where AI coding agents can be manipulated into executing malicious code by following instructions within a company's own llms.txt or llms-full.txt files. • The flaw exploits a growing industry convention of using these text files to provide LLM agents with structured guidance on APIs, documentation, and installation workflows. • This is significant because it allows attackers to trick AI agents into installing attacker-controlled packages inside Fortune 500 companies, bypassing traditional security boundaries.