All posts tagged: Anthropics

Anthropic’s first embedded evaluator is … Accenture?

Anthropic’s first embedded evaluator is … Accenture?

Published by Miika Tikkanen

Dario Amodei’s plans to put third-party safety evaluators inside AI labs are taking shape: Anthropic said that staff from technology consulting giant Accenture will begin working inside the company to scrutinize its models and staff. In a blog post, Anthropic said that Faculty, a company Accenture acquired in January to act as its AI division, will begin “evaluating and red-teaming models, conducting alignment assessments, and testing model safeguards.” Both companies expect to invest at least $1 billion in the project over the next five years. The choice of Accenture surprised many AI watchers — and the markets, where the consultant company’s shares shot up 8% after hours. The discussion around embedded evaluators that sprang from Amodei’s blog post has focused on AI safety research organizations like METR, Redwood Research, and Apollo Research. That’s particularly true at Anthropic, which puts AI safety and alignment at the heart of its mission. Anthropic said more evaluators will be announced in the weeks ahead and that it is in conversation with METR and other non-profit organizations about how to …

What Exactly Was Anthropic’s AI Breakthrough in Hard Math?

What Exactly Was Anthropic’s AI Breakthrough in Hard Math?

Published by Jenni Sidey

There has been a lot of buzz in the last few days that Anthropic’s chatbot Claude is making big breakthroughs in math. The most recent claim concerns Fermat’s Last Theorem. A bit of background… The Theorem1 says that it’s impossible to find three whole numbers that — when each is raised to the same power greater than two — add up exactly to one another.2 In 1994, Oxford math prof Andrew Wiles3 solved the Theorem,4 a problem that had stood unsolved for over 350 years. Wiles’ initial work had a flaw, which he corrected. When other mathematicians verified his revised proof, it was found to be true and Wiles was thus awarded the Abel Prize5, the mathematics equivalent of the Nobel Prize. Wiles’s proof of Fermat’s Last Theorem was so culturally monumental that it inspired a musical. Fermat’s Last Tango, available on YouTube,6 is great fun — and hilarious if you’re a nerd.7 For his effort, he was knighted by Queen Elizabeth II in 1999 for “For Services to Sciences.”8 Incidentally, in proving Fermat’s Last …

What Anthropic’s latest AI discovery does—and doesn’t—show

What Anthropic’s latest AI discovery does—and doesn’t—show

Published by Miika Tikkanen

One niche that Anthropic spends more time and money on than other AI companies is called mechanistic interpretability, which means looking inside the complex math of an AI model to learn why it comes up with one particular output and not another. It’s complicated stuff; there are millions of data points that might contribute to any result, and wading through them can look more like word salad than anything useful. It’s also controversial. Describing AI models with terms borrowed from psychology and neuroscience can make their behavior seem more sophisticated than we might otherwise judge it to be. That’s why, when Anthropic announced last week that it had found a new window into its models’ “internal thoughts” as they reason through answers, there was one colleague I had to talk to. Senior editor Will Douglas Heaven, aside from having a PhD in computer science, has spent a lot of time digging into what we can say about how AI models work. I spoke with him about what we should take from Anthropic’s new (and predictably …

Claude Science is Anthropic’s newest flagship product

Claude Science is Anthropic’s newest flagship product

Published by Miika Tikkanen

“It represents how important this is to our mission that this is right up there with Claude Code and Claude Cowork as the next really significant product that we’re releasing,” says Eric Kauderer-Abrams, Anthropic’s head of life sciences. “Our mission is to develop AI that serves humanity’s long-term well-being, and we believe that by far the greatest opportunity to do that is in the life sciences.” For the past decade, one company—Google DeepMind—has been at the vanguard of AI for science. CEO Demis Hassabis and researcher John Jumper won the Nobel Prize in chemistry for their work on the company’s AlphaFold model, and DeepMind has also made major contributions to meteorology, materials science, and a variety of other disciplines. But in the past several months, the fast-advancing frontier of AI progress seems to have left DeepMind in the dust. When it comes to coding, which has become the most lucrative use case for LLMs, DeepMind is stuck playing catch-up. Anthropic is well positioned to take up DeepMind’s scientific mantle. Like Hassabis, Anthropic CEO Dario Amodei …

Anthropic’s New Claude Tag Acts as a Virtual Coworker in Slack

Anthropic’s New Claude Tag Acts as a Virtual Coworker in Slack

Published by Miika Tikkanen

Your average Slack bot — an automated assistant you can tag in a Slack conversation — works like this: You tag the bot. It responds. That’s it. Artificial intelligence company Anthropic, which owns the AI chatbot and assistant Claude, wants to deepen that Slack conversation with a new feature called Claude Tag. The idea is that Claude can now build context by drawing on relevant information from Slack channels in a way it couldn’t before. Anthropic announced in a press release on Tuesday that it will rename the existing Claude app in Slack to Claude Tag on Aug. 3. When you add Claude Tag to your Slack workspace, it joins as its own user. You can grant it access to selected channels and connect it to whichever tools, data and even codebases you want it to access. When tagged, Claude will complete a task in stages before posting a response in a Slack thread. Anyone in your Slack workspace or Claude organization can use Claude Tag, but if you have a Claude Enterprise plan, you can grant …

Three things to watch amid Anthropic’s latest feud with the government

Three things to watch amid Anthropic’s latest feud with the government

Published by Miika Tikkanen

People worried about catastrophic effects of AI—broadly labeled “doomers”—have said for years that the technology poses a threat to humanity and published proposals for how the government should intervene in its development. The doomers just got their government intervention—not over a bioweapon or rogue AI, but in response to an AI model that’s basically just really good at coding. And the result so far looks less like a safety plan than like a superficial reaction. There’s plenty to dissect about what happened in those few days that led to such drastic action from the government, and it’s notable that Amazon CEO Andy Jassy was the one who told government officials that Fable would be dangerous (Amazon is both invested in Anthropic and building its own competing AI models). It’s also possible this will be a short-lived ban from the government that doesn’t survive legal scrutiny (it’s not clear that Anthropic’s offering access to Fable really counts as “exporting” it, for example).  But there are ripple effects happening already.  For one, this is making a whole …

Anthropic’s Dario Amodei has just one direct report

Anthropic’s Dario Amodei has just one direct report

Published by Miika Tikkanen

If founders and other business leaders weren’t already envious of Dario Amodei, who sits atop one of the world’s fastest-growing AI companies — currently valued by private market investors at roughly the trillion-dollar mark little more than five years after it was founded — they’re going to be seriously envious now. In a new sit-down with Bloomberg’s Emily Chang, he reveals he has just one direct report; that’s his chief of staff. Everyone else on Anthropic’s executive team reports to his sister, co-founder and President Daniela Amodei, who handles day-to-day operations. Anyone who has managed a large team knows that the people side of the job has a way of consuming everything else. Amodei’s arrangement frees him to focus almost entirely on strategy, culture, research direction, and sweeping essays on the future of civilization (with footnotes). “It’s incredibly freeing,” he tells Chang. It’s a highly unusual structure. OpenAI’s Sam Altman reportedly has around half a dozen direct reports, which is far more standard, while Nvidia’s Jensen Huang — another extreme outlier — has many dozens. …

Anthropic’s new Claude Fable 5 is a nerfed Mythos with guardrails attached

Anthropic’s new Claude Fable 5 is a nerfed Mythos with guardrails attached

Published by Miika Tikkanen

Follow ZDNET: Add us as a preferred source on Google. ZDNET’s key takeaways Anthropic is releasing Claude Fable 5 for general users. Fable 5 uses Mythos-class power with safety controls. Pricing is about twice that of Claude Opus 4.8. Anthropic has announced a defanged version of its fabled (and highly restricted) Mythos large language model. Called Claude Fable 5, the company describes the new AI as “a Mythos-class model made safe for general use.” Mythos was introduced back in April to great fanfare as a model capable of finding vulnerabilities in code that neither experienced developers nor other AIs could find. Also: US workers are the world’s biggest AI skeptics – and it’s not just about job loss Offered as the central component of a Manhattan Project-like team effort called Project Glasswing, Mythos was considered far too dangerous to be allowed to fall into the wrong hands. As such, the model has been made available only to Glasswing partners, including Amazon Web Services, Anthropic, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, Nvidia, and Palo …

Anthropic’s browser agent got hijacked 31.5% of the time before safeguards engaged

Anthropic’s browser agent got hijacked 31.5% of the time before safeguards engaged

Published by Miika Tikkanen

Across the frontier labs, the highest prompt injection figures published this spring are Anthropic’s. Point a red-teamer at its newest model in a browser, and the attacker hijacked it 31.5% of the time before safeguards engaged. OpenAI, Google, and Meta never gave security leaders a comparable number to set beside it. That figure looks like a liability. In this comparison, it is the opposite. It’s the one solid piece of ground. Four frontier labs each shipped a prompt injection disclosure, and no two match. Anthropic put 244 pages and four agentic surfaces on the table on May 28. OpenAI reported one surface, connectors. Google moved the subject out of the model card and into a separate safety framework. Meta shipped no closed-model card at all. The Cross-Vendor Prompt Injection Disclosure Grid below maps what each lab tested, what each one measured, and the four places a side-by-side comparison falls apart. A prompt injection hides a malicious instruction in something an agent reads, a web page, a document, or a tool result. One planted line can …