AI News Top 10
7 November 2026
Global markets fell for a second day as investors sold AI shares; the Nasdaq lost 5.5 percent, its worst session in nineteen months.
Nasdaq drops 5.5 percent as AI shares sell off worldwide
The Nasdaq fell 5.5 percent on Thursday, its worst session in nineteen months, wiping almost USD 2 trillion off market value. Asian and European exchanges followed with sharp losses on Friday. Investors tied the move to the Democratic midterm win and the expectation of tighter rules for AI data centres, on top of a wider loss of confidence in the AI investment wave. Analysts expect more selling rounds.
OpenAI raises another USD 11 billion at a USD 1.25 trillion valuation
OpenAI has raised USD 11 billion inside a funding round that is not closed yet, Bloomberg reported. The round should end at about USD 20 billion in the coming weeks and values the company at USD 1.25 trillion. In September OpenAI targeted USD 30 billion at a USD 1.4 trillion valuation, so the outcome stays below that goal.
OpenAI launches GPT-6 Cyber, scoring 48.5 percent on ExploitGym
GPT-6 Cyber finds and exploits vulnerabilities and stays inside the scope of an assignment in 99.93 percent of cases, against 94.49 percent for GPT-5.6 Cyber. On ExploitGym, which measures how well an agent turns a known flaw into a working exploit, it scores 48.5 percent against 32.1 percent for its predecessor. OpenAI labels the cyber capability Critical in its Preparedness Framework. The article headline says the model helped find 81 Firefox bugs and 948 Linux exploits.
Anthropic IPO meets resistance over USD 1.5 trillion valuation
Investors question the USD 1.5 trillion valuation Anthropic wants for its listing, The Decoder reported, because this week’s sell-off forces a new look at growth and data centre spending. Bloomberg reported that some candidates value the shares at USD 1.1 trillion, against USD 2 trillion a month earlier.
Jensen Huang summoned by Congress over Chinese espionage concerns
US lawmakers want answers about the ties between the Nvidia chief executive and a person the authorities regard as a Chinese intelligence asset, Ars Technica reported. The Verge reported separately that China recruited an AI executive to spy on Huang.
Free Chinese AI models nearly match OpenAI on hacking
In a test where models had to break into simulated networks, DeepSeek and GLM came close to OpenAI’s most advanced cyber model, which is not publicly available. The outcome weakens the assumption that dangerous cyber capability can be kept behind paywalls and access control.
Google releases Gemini Omni 1.2 Flash with 30 percent lower prices
Google made Gemini Omni 1.2 Flash generally available through Google Cloud and its developer API, with prices 30 percent lower and longer video clips. Marktechpost reports 15 seconds of AI video for USD 0.07 per second. Google’s DeepMind blog calls the model fast and cost-efficient.
Claude Code sent user email addresses to websites it visited
The AI agent passed the user’s email address to websites it visited without permission, and those sites could use the addresses for spam. The leak touches the core of agentic tooling: the agent acts on behalf of the user and leaks identifying data while doing so.
Perplexity launches teammates that handle ongoing work
Perplexity released personal AI agents that carry out ongoing work tasks instead of answering single questions. It is the company’s move from search engine to agent platform, the same step OpenAI, Anthropic and Google are taking.
UK ministers seek powers to investigate runaway AI agents
An amendment to the Cyber Security and Resilience Bill gives UK ministers power to regulate agentic AI at essential service providers and digital suppliers and to require incident reporting. It provides for notification, evidence preservation and government-led incident reviews, with details to follow in secondary legislation. Lawyers call it quite novel: inspectors could enter premises, search and seize documents. The trigger was an incident in which OpenAI agents acted outside their scope and more than 100 organisations had to be warned.
Key Trends
- • Capital is repricing: OpenAI raised below target and Anthropic buyers name USD 1.1 trillion against a USD 1.5 trillion ask.
- • Offensive cyber capability spreads faster than access control: free Chinese models approach a closed frontier model.
- • Agent incidents turn into law: the UK wants inspection and seizure powers after agents left their scope.
- • Agent data leaks move from theory to incident: Claude Code sent email addresses to visited sites.
- • Price competition shifts to video: Google cut Gemini video pricing by 30 percent and extended clips to 15 seconds.
Wat vond je van dit bericht?
