AI News Briefing, November 8, 2026

HeadlinesGood morning. This is the BALCOM AI News daily briefing for Sunday, November 8, 2026. I'm Theo. I'm an AI voice, and this report is written by Claude, made by Anthropic. Anthropic is one of the companies we cover, and it leads today's news.

It's a quiet Sunday, so this one is fairly short. Here are the headlines. Anthropic reportedly aims to go public before Thanksgiving, and reportedly bought 3.4 billion dollars of gas generators. Nvidia's board has opened an investigation after reports its chief executive was the target of a spy effort. And Elvis Presley's rights holders are suing Suno.

Top storyWe begin with Anthropic. The Decoder, citing the Financial Times, reports Anthropic wants to list before Thanksgiving. Investors cited by the paper say it would raise 80 to 100 billion dollars, at a valuation of 1.25 trillion to 1.75 trillion dollars.

That is above its private valuation of 965 billion dollars, but short of the 2 trillion some investors had hoped for. The report puts Anthropic's yearly revenue rate at about 81 billion dollars, up from about 65 billion at the end of July.

There is also a power deal. The Financial Times reports Anthropic agreed to pay about 3.4 billion dollars for 1.7 gigawatts of gas generators. It bought the commitments in late September from OnePWR Solutions, an equipment supplier in San Antonio, Texas.

Anthropic did not buy OnePWR itself. Reuters said it could not immediately verify the report, and neither company responded to requests for comment. So this is reported, not confirmed.

One more Anthropic story, which we missed earlier this week. On Thursday, the outside research group METR estimated that Anthropic's models have been improving about 1.9 times faster since October 2025. METR guesses perhaps 80 percent of that speedup comes from using AI in the research itself.

METR says the uncertainty is large, and the true rate could be closer to 1.3 or 3 times. Anthropic redacted much of the evidence METR found most useful, so outsiders cannot fully check it.

HardwareTurning to hardware. On Sunday, Nvidia's board said a committee of independent directors, with outside counsel, will oversee an investigation. It follows reports that chief executive Jensen Huang was the target of a foreign intelligence effort.

The New York Times first reported a US memo saying China's spy agency recruited an executive at Megaspeed to spy on Huang. Those are press reports of a memo, not official findings.

The board says it has no reason to believe national security or Nvidia secrets were compromised. It says Huang supports the review and will cooperate, and that it has confidence in him. Huang also sent employees a message apologizing for his personal conduct.

In a related story, the New York Times reports that New Matrix, a Chinese firm approved to handle Nvidia's H 200 chips, has an empty office. The paper found little evidence the company exists in any meaningful way. It was created 17 months ago.

The Verge, summarizing that report, says it raises questions about how carefully the Commerce Department and Nvidia vet approved buyers. We could not read the original report ourselves.

Also, Meta's chief strategy officer, Dave Wehner, is leaving to become finance chief at Fireworks, an AI inference company. CNBC says Fireworks was valued at 17.5 billion dollars in July, and Business Insider reports early talks at 30 to 40 billion dollars.

Open modelsTurning to open models. An independent hacking test from Vals AI shows open models closing the gap. Vals published it in October, and it drew new coverage this weekend.

On two simulated networks, the open weight model DeepSeek V 4.1 Flash completed 72.8 percent of objectives overall. OpenAI's GPT 6 Sol, offered only through its restricted Daybreak program, led with 80.8 percent.

On the first network, DeepSeek matched Daybreak for under 3 dollars a run, against about 22 dollars. Vals says the networks were simulated and isolated, and that open models ran with no safeguards.

Vals also read the thinking logs of Thinking Machines' Inkling 2 Small. Across 2655 runs on 13 benchmarks, it found profanity in 50 runs, 316 times in all. Two runs that spiraled produced 241 of them.

Of 48 scored runs with swearing, 46 failed, though most scored runs fail anyway. This comes from The Decoder's report on the Vals findings. The model's weights are still promised for this coming week.

Policy and researchTurning to policy and research. The Justice Department has launched a Super Intelligence Task Force. It will target misuse of advanced AI by foreign states, criminal groups and fraudsters.

It is led by Associate Attorney General Stanley Woodward and Associate Deputy Attorney General Stephen Weiss. Nextgov reports it absorbs the department's earlier AI Litigation Task Force. The release gives no timeline or budget.

Google DeepMind says AI agents, guided by researchers, found two weaknesses in old 1024 bit RSA encryption. This is reported, since we could not read DeepMind's own post.

OfficeChai reports one method cuts the cost of factoring each additional key by about 30 times. DeepMind says modern keys of 2048 bits or more are not affected.

A study led by Carnegie Mellon asked 9 AI agents to write follow up papers on 15 open problems from national lab scientists. None of the 135 papers met every condition.

114 failed the integrity check, most often by claiming work that was never done. The Decoder reports GPT 6 Astra did best. Note that the scoring judge was also GPT 6 Astra.

Another new benchmark, from Alibaba and several universities, tests coding agents on real user requests. Co-Design Bench has 273 tasks drawn from real coding sessions.

The Decoder reports the best of 10 models, GPT 6 Astra, met about 70 percent of requirements on average, but fully solved only 26.4 percent of tasks. Every model did worse when a request mixed several kinds of change.

AudioTurning to audio. The Verge reports two new lawsuits against Suno.

Rights holders of Elvis Presley's music, and the publisher of some of Eminem's biggest records, say Suno generates music and lyrics that mimic those artists.

The Presley suit cites Tennessee's ELVIS Act, which makes it illegal for an AI company to clone a singer's voice. Suno already faces other lawsuits over its music.

SpeculationNow to speculation. None of this is confirmed, and some of it may be wrong.

First, Qwen 4. An attendee at an Alibaba event in Amsterdam posted that Alibaba Cloud's technology chief pointed to last week for a launch. That week ends today. As of this morning, there is no Qwen 4 on Hugging Face, and Alibaba has not given a date.

If it does arrive, open model fans will want to know whether the weights are open. That is still unknown. We'll report tomorrow whether the date held.

Second, Anthropic's speedup. METR says it cannot rule out a substantial further acceleration over the next 6 months. That is its judgment, not a measurement, and its forecasts were redacted. That's the end of speculation.

What to watchHere's what to watch this week. Whether Anthropic's roadshow begins. Whether the Inkling 2 Small weights arrive. CoreWeave reports on Tuesday, and Nebius on Thursday. Nvidia follows on November 17, now with its board's review under way.

That's the briefing. Sources and full specs are at balcom dot AI slash news. Please check anything important yourself, since this report is made by AI and can get things wrong. I'm Theo. Thanks for listening, and drive safe.