Anthropic's CEO Breaks His Silence: We Never Wanted to Ban Open Models

After a week spent cast as the villain of the open-weights fight, Anthropic CEO Dario Amodei published a post on July 27 to rebut the charge head-on: "Anthropic has never advocated for a ban on open-weights models." The company had conspicuously declined to sign the industry letter defending open models (D.A.D., July 25), and critics—led by former White House AI adviser David Sacks—had accused the big closed labs of wanting Washington to stamp out their cheaper open-source competition (D.A.D., July 21). Amodei's answer is that he opposes a blanket ban too, but for different reasons than the letter's signatories: a ban "would not address my most serious national security concerns" and "would protect US AI companies from competition, but that has never been my goal." Open models without dangerous capabilities, he writes, are "a public good," and barring US businesses from using Chinese ones does nothing to stop genuine bad actors, "because bad actors are unlikely to be legitimate US businesses."

Instead, Amodei lays out three narrower measures he says Anthropic has consistently backed: keep advanced chips and chipmaking gear out of authoritarian hands (and crack down on smuggling), since without US chips China "cannot build more powerful models than the US"; crack down on "industrial-scale distillation"—the compute-cheap copying at the center of this month's Kimi fight (D.A.D., July 23)—which he says can bring China's models "within a few months" of the US frontier; and require mandatory pre-release safety testing of all sufficiently capable models, open or closed, for cyber, biological, and alignment risks. His real worries, he writes, are two "nightmare scenarios": that an authoritarian government—"the CCP is clearly the most capable threat"—builds AI more powerful than America's and uses it for "permanent military superiority" or deep repression; and that powerful models get misused for cyber or bio attacks. Open models, he concedes, do carry higher misuse risk than closed ones, since guardrails are hard to enforce and released weights "cannot be withdrawn"—but a ban on US business use, he argues, is simply the wrong tool for it. On the letter itself, he agrees open weights expand access and competition, yet rejects its assumption that they necessarily make safeguards easier or help defenders more than attackers—pointing to biology, where a capable model might help weaponize a pathogen faster than any defense could be mounted. That, he says, is a question for empirical testing, "not assumed in advance."

Why it matters: This is the fight's lone major-lab holdout finally explaining itself—and it scrambles the tidy "open vs. closed" story the week was told in. Amodei rejects the outright ban that hawks like Treasury's Bessent floated (D.A.D., July 23), yet endorses their crackdown on distillation; he defends open models as a public good, yet insists they can carry real misuse risks a feel-good letter shouldn't wave away. The effect is to drag the argument off the binary—ban or no ban—and onto a narrower, more technical menu: chips, distillation, and mandatory testing. Whether you take it at face value or read it as a more sophisticated version of the self-interest critics allege, it sharpens what's actually at stake: the real open question isn't whether open models are good or bad, but whether—on the most dangerous capabilities—wide access helps defenders or attackers more, something Amodei insists no one has yet proven either way.

Source: anthropic.com

Cohere Lets Businesses Chain Multiple AI Agents Into One Workflow

Cohere launched North Automations, a feature for its North enterprise AI platform that lets companies chain together multiple AI agents into coordinated workflows using plain-language instructions, with built-in spending controls and oversight. Cohere points to Gartner research suggesting many companies' AI agent projects underperform because agents work in isolation rather than together, and cites an internal example: its own marketing team using the tool to pull campaign data from BigQuery automatically. The company also published a hypothetical case study aimed at regulated fields, showing how a wealth manager could automate client research, market monitoring, and compliance checks. No independent performance data was provided.

Why it matters: Most companies today have scattered, single-purpose AI agents that don't talk to each other—orchestration tools like this aim to turn them into an actual workflow, and Cohere is clearly targeting regulated industries like wealth management where compliance work is time-consuming and high-stakes.

Source: cohere.com

Claude Code Can Now Turn a Session Into a Live, Shareable Web Page

Anthropic added a feature called Artifacts to Claude Code, its AI coding assistant, that turns whatever the tool produces in a session into a live web page hosted on claude.ai—and shareable with a link. Instead of copying results out of a terminal, you ask Claude to publish, say, a walkthrough of a code change with the differences annotated line by line, a dashboard built from data the session already pulled, or a status board that fills in while a long task runs. The page opens in a browser and keeps updating in place as the work continues. Artifacts are private to your account by default; you can share them inside your organization on Team and Enterprise plans, or, on the individual Pro and Max plans, via a public link anyone can open. The notable new twist: these pages can now call "connectors"—Anthropic's live links to outside tools like GitHub or a company's own systems—each time someone opens them, so a shared dashboard shows current numbers instead of a snapshot, pulling (and even acting on) each viewer's data through that viewer's own accounts. What it isn't is a full app: each artifact is a single self-contained page with no backend—a capture of work, Anthropic stresses, not software you'd run a business on. The feature requires a paid plan and a signed-in claude.ai account.

Why it matters: This is a small, concrete instance of a shift D.A.D. has been tracking—AI that spits out disposable, bespoke software, not just text. Y Combinator's fall startup wish list asked founders to build almost exactly this: a "cloud for small software" to deploy and share the one-off tools AI now makes trivial to create, as easily as sharing a Google Doc (D.A.D., July 23). Anthropic just shipped a version of it inside its own coding tool. For a non-engineer, the practical payoff is that the output of an AI work session—a report, a dashboard, a review walkthrough—becomes a link you hand a colleague instead of text you paste, and with live connectors it can stay current on its own. The limits are the point, too: it's a page, not a product, built for showing and sharing work rather than running it—so the near-term use is faster, richer handoffs, not a replacement for the internal tools IT actually maintains.

Source: claude.com

Claude Code Can Now Browse and Click Through Real Websites for You

Anthropic gave Claude Code, its AI coding assistant, a built-in web browser that Claude itself can read and operate—opening documentation, navigating an issue tracker, filling out a form, or clicking through a checkout on a live site, not just testing the app you happen to be building. The Browser is a tabbed pane inside the desktop app; you can sign in to sites within it (including Google-style popup logins), and Claude reads and acts on pages using the same tools it already uses to check your own app. Anthropic wraps it in guardrails plainly aimed at nervous IT departments: the pane runs a clean browser profile with none of your saved logins or history, kept separate from your personal browser and identity; safety classifiers review every click or keystroke Claude makes on an outside site and can force a permission prompt; each new site needs its own one-time approval; and Claude won't buy anything, create accounts, or solve CAPTCHAs without you. Administrators can restrict which sites are reachable, strip Claude of the ability to act on external pages while still letting people browse, or switch off external browsing entirely. For work that should happen as the logged-in you, Anthropic points users to its separate "Claude in Chrome" extension, which does share your real browser session.

Why it matters: This nudges AI coding tools past a boundary—from acting only inside the software you're building to operating the open web on your behalf. Reading a vendor's documentation, reproducing a bug through a web app, pulling a figure off a live dashboard: tasks that used to mean leaving the tool and doing it yourself now happen inside the same session. It's the same agentic direction as Anthropic's broader push into "computer use," where Claude carries out multi-step tasks on your machine (D.A.D., July 22), and the guardrails—sandboxed profile, per-site approval, no autonomous purchases—are the tell that the hard part isn't teaching AI to click, but making organizations comfortable letting it. The caution for readers is the one every agent story carries: an assistant that can navigate and fill in real websites is genuinely useful and, without a human watching the consequential steps, a fresh way for a confident mistake to act beyond your screen.

Source: claude.com

Dementia Therapists Get an AI Assistant for Memory Sessions

Researchers unveiled RemiAssist, an AI tool built to help therapists—not patients—run photo-based reminiscence therapy for people with dementia. The system organizes life-event photos into a structured map for session planning and offers real-time prompts to guide conversations and navigate emotionally sensitive moments. In a small field study of eight therapist-patient pairs, RemiAssist was linked to a 44% gain in planning efficiency and 54% longer conversations, along with better-handled sensitive exchanges.

Why it matters: It's an early example of AI aimed at augmenting a clinician's judgment in a delicate, human-centered therapy rather than automating the interaction itself.

Source: arxiv.org

Warning to Academics On Using AI For Data Analysis

Economists are increasingly feeding data through LLMs—nearly 6% of papers in top economics and management journals now mention them, and a July 2026 survey found 60% of economists call the tools essential to their work. A new paper warns this quietly undermines whether those results can be reproduced, because the same prompt sent twice can return different answers even with settings meant to remove randomness. To show the stakes, the authors ran a classic finance task 200 times: they had an OpenAI model score the sentiment of corporate annual reports (10-K filings), then regressed stock returns on those scores—holding the filings, prompt, and model fixed across every run. The answers wobbled enough that in roughly 5% of runs a researcher would report a "statistically significant" finding that, rerun, cleared the bar only by luck. Turning the "temperature" dial (a randomness setting) to zero helped but didn't fix it: silent model updates, server batching, hardware differences, and the "mixture-of-experts" routing that shunts your request to whichever specialist component has room all inject variation an online user can't control. Worse, the newest reasoning models from OpenAI (GPT-5.6) and Anthropic (Sonnet 5, Opus 4.8, Fable 5) have removed the temperature control altogether. Only a small open-weight model run locally on fixed hardware reproduced its own answers every time. The authors' bottom line: treat LLM output as "draws from a distribution rather than fixed measurements," and document the exact model version, date, prompts, and raw outputs so a study can even be checked.

Why it matters: As academic and business research leans on AI to classify documents, score sentiment, and extract numbers, this is a caution that those outputs aren't fixed facts but statistical estimates with error bars—and that rerunning the identical analysis can move the answer, occasionally flipping whether a result clears the threshold for significance. For anyone citing AI-generated analysis as evidence—in a paper, a market model, or a compliance report—the practical takeaway is to log the model version and settings and, where the stakes are real, run the job several times to see how much the answer drifts rather than trusting a single pass.

Source: arxiv.org

A Way to Prove AI Can't Misuse Your Mental Health Data

Digital phenotyping tools infer mental health conditions from behavioral data like spending patterns and phone use, raising obvious privacy and consent concerns. Researchers built a formal system that encodes ethical rules—like when data can be used—as logical constraints, then used automated verification software to prove those rules can't be violated within the model. Applied to a case combining financial and mental health data, the approach checks compliance continuously rather than through after-the-fact audits or paperwork. There is no real-world performance data yet—this is a proof-of-concept.

Why it matters: As AI systems increasingly infer sensitive personal states from mundane data like transactions, this kind of built-in, verifiable guardrail could become a compliance requirement rather than a research curiosity.

Source: arxiv.org