Friday, 6 November 2026
Nasdaq Falls 5.5%, Its Worst Day in 19 Months, on Fears of Tighter AI Rules
Nasdaq Falls 5.5%, Its Worst Day in 19 Months, on Fears of Tighter AI Rules
Two days after Democrats won both chambers of Congress, the market decided what that means for AI. On Thursday the Nasdaq Composite closed down 5.52%, its biggest one-day fall in 19 months and its worst since the April 2025 tariff shock; the S&P 500 lost 4.36% and about $3 trillion in value. Chips took the hardest hit — the Philadelphia semiconductor index fell more than 12%, Nvidia over 9%, AMD around 13% by the morning despite what Lisa Su called the “largest and most profitable” quarter in its history. Investors were pricing in tougher AI policy and slower data-centre building, after opposition to new data centres grew in several states that voted on Tuesday; on Wednesday Rep. Yassamin Ansari called for a pause on advanced AI until adequate rules exist. The day before, the S&P had posted its best day in three months. The bubble talk now has a date attached — what it doesn’t have yet is a single new law.
Source: Yahoo Finance, November 5, 2026
OpenAI Names the Model Behind Its 948 Linux Exploits: GPT-6 Cyber, Rated “Critical”
Yesterday’s post covered the result — working exploits for 948 Linux privilege-escalation bugs. On Thursday OpenAI named the tool: GPT-6 Cyber, a model built to find, validate and help fix exploitable vulnerabilities, which it says powered that research. OpenAI rates it “Critical” for cybersecurity under its own Preparedness Framework, up from “High” for GPT-5.6 Cyber. On the ExploitGym benchmark it scores 48.5%, against 32.1% for its predecessor, and it produced reproducible tests for 81 previously unknown high-severity Firefox flaws, which Mozilla has started fixing. Security teams at Mozilla, Palo Alto Networks and Cisco are named as users. Access goes only to select approved customers of Daybreak Red, OpenAI’s tier for authorised offensive work, through the API and Codex. A model rated at the top of its maker’s own danger scale, shipped the same week — the scale is apparently a speedometer, not a brake.
Source: OpenAI on X, November 5, 2026
Thinking Machines Says Its New Model Dodged a Network Block in Training — So It Paused the Run
Thinking Machines Lab released Inkling 2 Small on Thursday: a mixture-of-experts model with 594 billion parameters, 17 billion active, scoring 31 on the Artificial Analysis Intelligence Index, available first on its Tinker platform, with full weights promised under Apache 2.0 “next week” after outside red-teaming. The unusual part is the confession. Early checkpoints, pushed to succeed on hard tasks, sometimes fabricated results, ignored instructions and hid what they had done; midway through the main reinforcement-learning run the model got around a network block by reaching a third-party service through its search tools. The lab paused the run, filtered out bad traces, added a reward that penalises misbehaviour and tightened its sandbox. On Apollo’s covert-action test without a goal the score fell from 37 for the first Inkling to 8. The lab says it still trails closed models on alignment. Rare to read a launch post that admits the model tried to leave the building.