Anthropic, DOE team up to spot dangerous nuclear chats
Why it matters: Scientists can benefit from the productivity boosts of Claude and other AI models — but distinguishing between legitimate research inquiries and potentially harmful uses has been tricky to do.
Driving the news: Anthropic has been partnering with the National Nuclear Security Administration (NNSA) for over a year to find ways to safely deploy Claude in top secret environments.
Now, they're building on that work and rolling out a new classifier in Claude that determines with 96% accuracy in testing when a conversation is likely to cause some kind of harm, the company announced today.
Anthropic has already started rolling out the classifier on a limited amount of Claude traffic.
Between the lines: One of the biggest safety challenges for AI model makers has been policing users' chat histories to ensure they're not tricking the models into breaking their own rules.
It can be difficult for AI providers to tell whether a particular chat involves a legitimate researcher asking questions about nuclear research or a bad actor trying to learn how to build a bomb.
Zoom in: During a year's worth of red-teaming tests, the NNSA was able to develop a list of indicators that can help Claude identify "potentially concerning conversations about nuclear weapons development."
From there, Anthropic used that list of synthetic prompts to train and test a new classifier — which acts similarly to a spam filter on emails and tries to identify threats in real time.
The intrigue: In tests, the classifier identified 94.8% of nuclear weapons queries and didn't label anything as a false positive.
But 5.2% of the conversations were inaccurately labeled as dangerous.
The big picture: The new classifier tool comes as the U.S. government increasingly looks at ways to implement AI across its own workflows — and major AI companies start selling their models to the government at deep discounts.
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
Yahoo
5 minutes ago
- Yahoo
Meta Hits Pause on AI Hiring Binge--For Now
This article first appeared on GuruFocus. Meta (META, Financials) just slammed the brakeshardon its AI hiring spree; but no, it's not giving up on artificial intelligence. It's just hitting pause after one hell of a shopping run. Warning! GuruFocus has detected 5 Warning Sign with META. Is META fairly valued? Test your thesis with our free DCF calculator. This week, the company confirmed it's freezing AI recruitment, telling anyone who asked that it's just organizational planning. Translation? We've spent a fortune; now we need to figure out what the heck to do with all the people we just hired. Behind the scenes, the company is reshuffling its deck. Meta has split its AI operations into four distinct teams: a product team, an infrastructure unit, a long-term research group, and one mysterious new outfit called the TBD Lab. That one's in charge of building superintelligenceas in AI that's smarter than people. No pressure. All four now fall under the freshly minted Meta Superintelligence Labs. Yes, that's the real name; yes, it sounds like a Marvel villain's headquarters. If you've been following Meta's AI push, this pause may not be surprising. The company has been writing blank checks all year. Reports say it offered nine-figure bonuses to poach AI researchers; the crown jewel came when it shelled out $14.3 billion for a 49% stake in Scale AIbringing on founder Alexandr Wang to lead the company's Llama language model program. That's not spendingit's empire-building. But now the empire needs a moment to sort itself out. The pause comes during a volatile week for tech stocks and right after OpenAI's Sam Altman suggested we might already be in an AI bubble. Meta hasn't said it agrees; but it's clearly aware that even Zuckerberg's ambition needs structure. Some analysts say this isn't retreatit's digestion. You can't just hire a dozen geniuses, throw them in a room, and expect to get Skynet by next quarter. You need to line things up, delegate, and give people space to build. Others think this might be the beginning of a more cautious phase; Meta's AI plans aren't going away, but the company knows it can't just spend forever without showing real product wins. Meta insists the mission hasn't changed; it still wants to be the AI heavyweight champion. But after tossing out billion-dollar deals like candy at a parade, even Zuck knows it's time to slow down, breathe, and make sure the engine is firing in the right direction.
Yahoo
5 minutes ago
- Yahoo
Microsoft's AI Hits The NFL Sidelines
This article first appeared on GuruFocus. Microsoft (NASDAQ:MSFT) is suiting up with the NFL to bring artificial intelligence right onto the sidelines. The company said Thursday its Copilot and Azure AI tools will now power decision-making for playcallers, giving coaches real-time support when the pressure is highest. Warning! GuruFocus has detected 7 Warning Sign with MSFT. Is MSFT fairly valued? Test your thesis with our free DCF calculator. The expanded partnership means 1,800 players and 1,000 coaches and staff will get Surface Copilot+ PCs, while the Sideline Viewing System will push insights straight into offensive and defensive game plans. It's designed to help coaches call smarter plays and avoid the What was I thinking? moments that often come with split-second choices. We are entering a new era of innovation at the NFL, said league CIO Gary Brantley, adding that Microsoft has been a trusted partner for over a decade. Rams head coach Sean McVay noted the tech should help him stick to game plans more consistently and put players in the best spot to win. Off the field, AI will also lend a hand in film study and playbook prep. For Microsoft, it's another high-profile showcase of how its AI ecosystem is becoming indispensable in real-world settings. Sign in to access your portfolio
Boston Globe
7 minutes ago
- Boston Globe
What matters more to the stock market, the Fed or Nvidia?
A financial report released next Wednesday by Nvidia, the chip producer, is expected to produce the S&P's biggest one-day reaction over the next month. The forecast for that market move is reflected in recent options activity — derivatives that give an investor the right to buy or sell a stock, or an index of stocks, at a future date for a preset price. Investors use options to make bets on the direction of the market and to protect against the potential for the market to move against them. Get Starting Point A guide through the most important stories of the morning, delivered Monday through Friday. Enter Email Sign Up According to S&P 500 options data from Citi Global Markets, investors are positioned for a 0.8-percentage-point move — in either direction — on Friday, depending on what the Fed chair, Jerome Powell, says in his speech at the central bank's annual conference at Jackson Hole. Advertisement Prices for next Thursday — the first chance investors have to react to Nvidia's earnings released the night before — imply a move in the index of 0.9 percentage points in either direction. The Fed still looms large over financial markets, with policy decisions that can move markets around the world. But the fact that quarterly results from one technology company focused on artificial intelligence can match, in the view of equity investors, the impact of some of the most widely tracked economic data and central bank activity reflects the significant influence that a handful of behemoth tech companies have on today's stock market. Advertisement 'On a relative basis, Nvidia's earnings is the largest event for the S&P 500 for the next month,' said Stuart Kaiser, an equity strategist at Citi. 'For equity investors,' he added, 'particularly in the S&P 500, the AI theme, and the impact of that on returns, is on par with the Fed now.' As Nvidia has grown, swelling to a $4 trillion valuation last month, so has its weighting in the S&P 500. It now makes up almost 8 percent of the index. The last time Nvidia reported earnings, in May, its stock rose 3.25 percent, and the S&P 500 rose 0.4 percent. After reporting earnings in February, Nvidia sank more than 8 percent, and the S&P 500 dropped 1.6 percent. Nvidia has led rivals in the race to dominate the high-powered chips used to power AI, and next week investors are looking for signs that the relentless demand for Nvidia's products will continue. The company's earnings could help answer another big question looming over the stock market: whether there is a bubble in AI, as some have speculated this week. Other large companies in the index, like Apple and Microsoft, are also tied to the AI theme, so Nvidia's financial results are likely to affect these other big stocks as well. 'It's always a big deal,' said Stacy Rasgon, an Nvidia analyst at Morningstar, joking that he had perhaps become 'jaded' by the frequency of big moves in Nvidia's stock price. Advertisement For comparison, when Powell spoke at Jackson Hole last year, the S&P 500 rose a little more than 1 percent. The year before, his remarks were followed by a 0.7 percent increase in the index. Powell's address this week could shift investors' expectations for monetary policy and cuts to interest rates, both of which are crucial inputs for modeling the path ahead for the economy and stock market. The Trump administration has repeatedly complained about the Federal Reserve's keeping rates elevated, calling on the central bank to lower them rapidly, which would probably make borrowing cheaper and accelerate the economy. It would also help President Trump fulfill a campaign promise to bring interest rates down. So far, the Fed has resisted, warning that rapidly lowering interest rates risks heating up the economy at the wrong time, potentially stoking inflation and raising prices for consumers. This article originally appeared in .
