Fei-Fei Li, the pioneering Stanford AI luminary and World Labs chief, is speaking at our sixth annual Fortune Brainstorm AI, December 7 to 8 in San Francisco.

So is Palo Alto Networks CEO Nikesh Arora, Accenture Song CEO Ndidi Oteh, and Apptronik CEO Jeff Cardenas.

And I can’t forget the founder-CEOs: Andrew Feldman of Cerebras Systems, Benham Neyshabur of Mirendil, and Jim Belosic of SendCutSend are also speaking. Mmmhmm.

We’re just getting started, of course—many more speakers to come. Wanna be in the room? Best to register before we reach capacity. (We always do.)

Today’s tech news below. —Andrew Nusca

Want to send thoughts or suggestions to Fortune Tech? Drop a line here.

Anthropic upgrades Claude Fable and Mythos to 5.1

Natalie Behring/Getty Images

Anthropic’s Fable and Mythos—the wide-release and limited-access versions of the same Claude model—have received point upgrades two and a half months after their original launch.

(“Controversial,” you’ll recall, because Mythos scared the dickens out of the American military-industrial complex with its supernatural software flaw-finding capabilities that, in the wrong hands, could lead to a tidal wave of crippling digital attacks.)

Like any good tech upgrade, Anthropic says the new versions are more capable and less costly than their predecessors. Fable 5.1, for example, “avoids shortcuts that result in poorer-quality work” and fixes “the root causes of software issues” yet costs up to 45% less for agentic work, its maker estimates.

The IPO-bound San Francisco company is so confident in its advancements that it believes they “add to the evidence that AI models will soon make important contributions to scientific discovery,” particularly for molecular design, computational analysis, and computational biology.

Still, it’s the risks of being so capable that keep people up at night. 

Anthropic says Mythos has safeguards in place to prevent it from creating chemical or biological weapons, stop it from hacking into systems it shouldn’t, and discourage it from accepting malicious agentic coding requests. It also has “anti-distillation mechanisms” in place to foil competitors’ attempts at copying its capabilities.

“AI models’ agentic capabilities have become much more powerful over the past two years,” the company wrote. “But … greater autonomy comes with new risks.” —AN

OpenAI to limit access to Astra model’s advanced cyber features

OpenAI is changing its model launch strategy as its technology becomes more powerful and the potential for its misuse grows.

That’s especially true following the July incident in which the AI models OpenAI was testing autonomously planned and executed a cyberattack against AI company Hugging Face.

OpenAI’s next model, Astra, comes out “soon,” the company recently said. Astra is expected to be substantially more capable than the company’s current frontier AI model, GPT-5.6 Sol.

But only a handful of partners will get access to the model’s most advanced cybersecurity capabilities as OpenAI works to balance helping companies prevent cyberattacks while not empowering attackers at the same time.

OpenAI says it is courting customers for so-called defensive cybersecurity. It sees these sales as a critical revenue stream and a main priority for its new chief revenue officer Dali Rajic.

A small group of “alpha testers” with full access to Astra’s cybersecurity capabilities includes “individuals and organizations that are responsible for protecting critical digital infrastructure and, broadly, critical infrastructure,” the company said. (It declined to name them.)

OpenAI said it will monitor how the model performs among this small group and will expand access through its Daybreak Blue program once it is confident Astra has “the right calibration” and it can “provide defensive benefits while reducing the potential for misuse.” 

After all, Astra’s release was already delayed by several weeks in the wake of the Hugging Face incident. “We took extra time to make sure that what we’re launching is safe,” a spokesperson said. —Emily Forlini

Tax tokens to fund jobs? A new bill tackles AI-driven unemployment

If artificial intelligence indeed brings widespread unemployment, one group of lawmakers wants AI companies to foot the tax bill. 

A new U.S. House of Representatives proposal would impose an excise tax on major AI companies and automatically raise the rate if unemployment climbs, funneling the money into creating jobs in areas from housing construction and infrastructure to child and elder care. 

“If Congress does nothing, the rise of AI could create the biggest wealth transfer in history from the bottom to the top,” said Rep. Sara Jacobs (D-Calif.) in a prepared statement. “If AI profits off human work, workers deserve job security and a share of those profits.”

Introduced earlier this month by Jacobs alongside representatives Greg Casar (D-Texas) and Valerie Foushee (D-N.C.), the bill proposes a bifurcated taxation: either tax the value of the tokens or tax revenue from AI services and certain transactions with affiliated companies, whichever yields the higher sum. 

The rates would start at 2% and 3% respectively when unemployment is 5% or less, and rise as unemployment increases. 

The bill is the most recent attempt in a concerted effort by both chambers of Congress (and by both Democrats and Republicans) to combat potential job displacement as a result of AI. 

Among their recent proposals: Studies of AI-related job displacement, required disclosure of AI-driven layoffs, a new excise tax on data centers, a tax on AI companies in part based on energy usage, the establishment of an AI sovereign wealth fund, and tax credits for expenses from retraining workers for AI. —Mia Osmonbekov

More tech

—Three quarters of Apple CEO John Ternus’s equity award for fiscal 2027 will vest based on Apple’s total shareholder return relative to other S&P 500 companies.

—World Labs unveils Atlas. Fei-Fei Li's AI startup shows off a multimodal world model that uses text, images, video, and 3D input to generate camera-controlled spatial output. 

—Dell shares jump 9%. Q2 revenues of $46.97 billion handily beat estimates; so does a full-year 2027 revenue forecast of $192 billion.

—American service members will gain access to military versions of OpenAI’s ChatGPT and SpaceX’s Grok.

—Sony faces a class action lawsuit over whether its gaming customers were misled in expecting permanent access to PlayStation titles they purchased.

—Palo Alto Networks acquires Console, the San Francisco firm behind the AI-powered IT service management platform of the same name.

—HP is looking for a new rhythm in Asia via its Garage 2.0 startup incubator initiative.