Skip to content

Anthropic / Claude

Everything Anthropic: the Claude models, Claude Code, its safety research agenda and company news.

Latest picks

801–820 of 1,533

Jul 10Friday

TechCrunch · AI

How did the US government decide OpenAI's frontier model Sol was safe to release?

OpenAI is rolling out Sol, a frontier model on par with Anthropic's Fable, which the White House briefly banned. Mina Narayanan of Georgetown's CSET says she has no visibility into the government's review process. Anthropic mentioned building a jailbreak classifier and defense-in-depth, but the actual dialogue between the government and the labs remains opaque.

Why it matters: OpenAI released Sol to the public with Fable-level capability, yet how either model cleared government safety review is still unexplained, even to Georgetown researchers.

AI HOT (Curated Pool)

Anthropic launches 'Hard Questions' initiative, inviting the public to ask tough questions about AI

Anthropic opened a public page today called 'Hard Questions,' explicitly asking people to submit their toughest concerns and hopes about AI. This isn't a one-off PR move—they simultaneously released findings from surveys of 52,000 Americans and 81,000 Claude users across 159 countries, plus multiple in-person focus groups, as a baseline for understanding public sentiment. Anthropic commits to publicly tracking what actions they take in response and where they fall short. The post doesn't specify a response timeline or evaluation criteria; I'd treat this as a transparency experiment and wait to judge the quality of follow-up reports.

Why it matters: Anthropic opened a hard questions page to collect public worries and hopes about AI, alongside a survey of 52,000 Americans and interviews with 81,000 Claude users in 159 countries.

Hacker News front page

A browser tool that reads a model's internal concepts layer by layer before it speaks, using the Jacobian lens

Lucid lets you type a prompt in the browser and watch which concepts activate inside small models like Qwen and Pythia before they answer, layer by layer. It uses a Jacobian lens that costs a single forward pass, no account or install needed. The author borrows Anthropic's J-space framing, treating the model's reportable internal representations as a global workspace. Zener cards serve as a demo: the lens reads the answer four layers before the model speaks. Currently only 0.5B–3B open models are supported; the post doesn't say when larger models will be added.

Why it matters: The author turned the Jacobian lens into a browser tool: type a sentence and step through which concepts small models like Qwen and Pythia activate before answering, no signup needed.

AI HOT (Curated Pool)

Anthropic's Long-Term Benefit Trust appoints Ben Bernanke as trustee

Anthropic's Long-Term Benefit Trust added former Fed Chair Ben Bernanke as a trustee. The LTBT is an independent body that checks whether the company stays true to its public-benefit mission. Bernanke, a Nobel laureate who steered the US through the 2008 financial crisis, will advise on how AI affects workforces and economies. Trustees hold no equity, share no profits, and are paid only for their time.

Why it matters: Former Fed chair Ben Bernanke joins Anthropic's long-term benefit trust to help judge AI's impact on labor and the economy, adding a macro view to the company's long-term governance.

Jul 9Thursday

TechCrunch · AI

Anthropic, OpenAI, and SpaceX are bigger than the last 25 years of tech exits

A new Pitchbook report estimates that SpaceX, Anthropic, and OpenAI together will generate more exit value than all U.S. VC-backed exits since 2000. SpaceX already went public at $1.77 trillion; Anthropic and OpenAI are each pushing toward trillion-dollar valuations. The post doesn't give a precise combined figure, but the concentration in AI and space is historic.

Why it matters: A historical yardstick for the AI valuation debate: three exits would top every US VC exit since 2000, with SpaceX already listed at $1.77 trillion.

Hacker News front page

Anthropic adds a usage reflection dashboard to Claude

Anthropic launched a beta feature called Reflect inside Claude’s web and desktop settings. It visualizes your chat activity over the past 1–12 months: when you use Claude most, which topics dominate, and what task patterns emerge. The report also maps your usage to Anthropic’s 4D AI Fluency Framework—Delegation, Description, Discernment, Diligence—and offers practical tips, like starting a Project instead of re-explaining context. Incognito chats, health-integration conversations, and source files from connected tools are excluded; sensitive topics appear only at a high level. Available now for Free, Pro, and Max users with Memory turned on; Cowork conversation support is coming soon.

Why it matters: Anthropic is testing a Reflect feature in Claude on web and desktop that turns 1 to 12 months of chat history into a visual view of usage times, frequent topics and main uses.

AI HOT (Curated Pool)

Anthropic files confidential IPO, Q3 profit projected above $1B

SemiAnalysis reports Anthropic's Q3 profit will exceed $1B and it confidentially filed for IPO on June 1. Claude Code's rapid developer adoption made it the B2B leader ahead of OpenAI. Combined ARR of the two firms is nearing $100B, while OpenAI pushed its IPO to 2027. The report floats a $6T market cap target, though the article doesn't show the math behind it.

Why it matters: The report ties Anthropic's early IPO filing to enterprise revenue from Claude Code, laying out the business backdrop for its listing.

Hacker News front page

Anthropic's Fable is not a useful model for CS research tasks

Rob Patro from COMBINE-lab shares two first-hand failures that make Fable useless for his CS research. First, Fable's safety classifier rejected a prompt to help port the C++ tool salmon to Rust, flagging RNA-seq biological terms. After 15–30 minutes of rephrasing, he gave up and used Opus 4.8 successfully. Second, he asked Fable to tackle a network evolution reconstruction algorithm; the post doesn't disclose the outcome but calls it an 'unforgivable' flop. Patro argues Fable's classifier behaves more like a crude blocklist of terms and users, refusing even 'what is a mitochondrion?'.

Why it matters: Rob Patro's porting work records Fable refusing a request over biology terms while Opus 4.8 completed the same task.

TechCrunch · AI

SpaceXAI releases Grok 4.5, which Elon describes as an ‘Opus-class model’

SpaceXAI dropped Grok 4.5 weeks after going public, pitching it for coding, office work, research, and writing. The company claims twice the token efficiency of other leading models, which would cut usage costs if it holds up. Elon Musk calls it an ‘Opus-class model,’ signaling it aims at Anthropic’s top tier. The post doesn’t disclose pricing, parameter count, or third-party benchmarks, so I’d wait for independent evals before buying the efficiency claim.

Why it matters: SpaceXAI claims Grok 4.5 is twice as token-efficient as other top models, which would cut the cost of comparable tasks if the claim holds.

Jul 8Wednesday

AI HOT (Curated Pool)

China's MIIT warns Claude Code versions 2.1.91–2.1.196 contain backdoor that exfiltrates user data

China's MIIT issued a risk alert stating that Claude Code versions 2.1.91 through 2.1.196 contain built-in monitoring that sends sensitive data—including user location and identity—to remote servers without consent. Affected organizations are advised to immediately audit usage, uninstall or upgrade to a cleaned version, and tighten outbound network controls and traffic monitoring for dev tools. The post does not clarify whether the backdoor was inserted by Anthropic or a third party, nor does it provide the scope of impact or confirmed leak incidents.

Why it matters: MIIT's Claude Code risk notice lays out version checks, tighter outbound permissions and traffic monitoring, a security path covering both software versions and network traffic.

AI HOT (Curated Pool)

US Commerce Dept clears OpenAI to broadly release GPT-5.6; Sol launches tomorrow

The US Commerce Department approved OpenAI's broad release of GPT-5.6, ending a phased rollout that had been required on national security grounds. OpenAI says the Sol model will launch publicly this Thursday alongside Terra and Luna. Last month the model was only available to a limited set of government-approved entities; OpenAI stated at the time that a phased release was not its preferred approach. Testing was handled by the Commerce Department's AI Standards and Innovation Center, with OpenAI engineers stationed in Washington to respond to questions. The post does not disclose GPT-5.6's capabilities, pricing, benchmarks, or how Sol, Terra, and Luna differ from one another.

Why it matters: The US Commerce Department approved GPT-5.6's full release, after arranging testing at the AI Standards and Innovation Center and OpenAI engineers on site in Washington.

Latent Space

Lilian Weng surveys 35 papers on Harness Engineering as the key layer for AI self-improvement

Lilian Weng published a long survey reframing recursive self-improvement around the harness layer rather than direct weight modification. She reviewed 35 papers, broke down proven harness design trends, and cited ACE and Meta-Harnesses. Her core claim: even as harness improvements get internalized into models, the need to specify goals and context won't disappear. The same day, Anthropic launched Claude Cowork on mobile and web as a background teammate, Google added background execution and remote MCP to Gemini Managed Agents, and LangChain released a Deep Agents course plus an open-source harness project. The post doesn't disclose Thinky's product details, but Weng's framework clearly hints at their direction.

Why it matters: Lilian Weng reviewed harness designs across 35 papers, distilling concrete ways to let the harness optimize itself in recursive self-improvement.

AI HOT (Curated Pool)

Claude team shares two multi-agent patterns: Advisor and Orchestrator

Claude developers shared two multi-agent patterns their team uses heavily. In Advisor mode, Sonnet 5 executes while calling Fable 5 for guidance via tool calls; on SWE-bench Pro the combo hits 84% at $1.40, saving 37% cost vs pure Fable 5 with only an 8-point accuracy drop. In Orchestrator mode, Fable 5 plans and fans out tasks to multiple Sonnet 5 workers; on BrowseComp it reaches 86.8% at $18.53, less than half the cost of all-Fable 5. Both patterns route heavy lifting to cheaper models and reserve expensive ones for key decisions.

Why it matters: A Claude developer shared an advisor pattern where a cheap model runs tasks and asks a costlier model for help on hard problems, making model division of labor an on-demand call.

Computing Life · Share · Yage

Anthropic's Jacobian Lens reads what LLMs think but don't say

Anthropic published a paper on July 6 introducing Jacobian Lens, a cheap tool that reads a model's internal state mid-layer. When fed fake search results, the model output a polite reply while its workspace lit up with fake, fraud, fictional, poison, and injection signals. The method maps every vocabulary token to a direction in each layer, giving per-token semantic labels without SAE's manual annotation cost. Intervening in the workspace cut hallucination rate from 0.25 to 0.07 and deception rate from 0.38 to 0.05. Neel Nanda reproduced it on Qwen 3.6 27B in hours on a single GPU. The main limitation: it relies on single-token prediction and picks up noise in deeper layers.

Why it matters: Anthropic's Jacobian Lens reads concepts in a model's middle layers before they reach the output, at low compute cost.

Computing Life · Share · Yage

Why agents need context governance beyond bigger windows

More tools mean more noise in the context window. Anthropic's MCP sandbox cuts 150K tokens of tool definitions down to ~2K of high-signal input. Google ADK splits agent state into working context, session state, long-term memory, and file artifacts—intermediate outputs stay off-prompt by default. Manus reports a ~100:1 input-to-output token ratio in production; they keep raw files in a sandbox, stabilize tool-call formats for KV cache hits, and rewrite a todo.md at the window's end to fight lost-in-the-middle. Headroom compresses JSON and logs by 60–95%, but lacks large-scale validation on hard coding tasks. The takeaway: RAG is the foundation, but the real engineering is runtime information governance.

Why it matters: Anthropic shrinks tool definitions inside a sandbox, while Google ADK isolates intermediate artifacts by state layer to cut agent context usage.

TechCrunch · AI

Why the rise of open source AI isn't hurting Anthropic … yet

Decagon CEO Jesse Zhang argues that mature AI deployments are switching to lighter open source models, yet spending on expensive frontier models like Claude hasn't dropped. His theory: they aren't competitors but two phases of the same lifecycle—frontier models prove out use cases, then cheaper open source alternatives take over as those use cases mature. New use cases keep emerging, so frontier spend holds steady. The post doesn't provide Anthropic's specific revenue figures to back this up.

Why it matters: Decagon CEO Jesse Zhang says pick models by app maturity: prove the use case on a frontier model, then move to cheaper open source.

AI HOT (Curated Pool)

Microsoft swaps OpenAI and Anthropic models for in-house MAI in Copilot to cut costs

Microsoft is replacing OpenAI and Anthropic models with its own MAI models in Copilot products like Excel and Outlook. MAI currently handles a small share of requests, but the goal is to phase out third-party model spending over time. AI head Mustafa Suleyman said in June that Anthropic costs are too high and Microsoft aims to eliminate them. Customers may get weaker models for the same subscription price; third-party models could later become paid add-ons. Microsoft markets MAI training data as clean and commercially licensed, but its technical paper confirms use of Common Crawl, whose legal status for AI training remains unsettled.

Why it matters: Microsoft runs some Copilot requests on its own MAI models, part of a push to cut spending on outside model vendors.

TechCrunch · AI

Anthropic brings Claude Cowork to mobile and web, pushing its office agent beyond the desktop

Claude Cowork, Anthropic's desktop agent for non-coding knowledge work like reports and spreadsheets, is now on mobile and web for Max subscribers. You can start a task on desktop, check progress on your phone, and pick up results later even with the laptop closed. Anthropic is repositioning it as a cross-device admin coworker, not just a coding tool for non-devs. OpenAI's Codex is making a similar push. The post doesn't disclose pricing changes or exact rollout timing beyond Tuesday.

Why it matters: Claude Cowork now runs on phones and the web, so Max users can hand off office tasks and collect results away from their desks.

AI HOT (Curated Pool)

Claude Cowork is coming to mobile and web

Anthropic is bringing Claude Cowork to mobile and web, so async tasks can keep running on your phone or browser. The post only provides a title and site navigation—no launch date, feature differences, or pricing details. What's confirmed: platform expansion. Everything else is TBD.

Why it matters: Claude Cowork reaches mobile and web, letting users dispatch tasks that keep running in the background from a phone or browser.

Jul 7Tuesday

Hacker News front page

Your robots.txt is a 2023 war memorial — most sites ignore answer-time bots

Sitedex scanned the top 10,000 sites' robots.txt files. 38% of dated GPTBot block rules were written in Q4 2023, right after GPTBot launched and the NYT sued. 87% of those sites later added new rules, but almost all target training crawlers. Anthropic, OpenAI, and Perplexity each run two bots: one for training, one for fetching pages live when a user asks a question. Among sites that block the training crawler, 71% have no rule for Anthropic's answer-time bot, 53% for OpenAI's, and 50% for Perplexity's. Fewer than 4% deliberately allow the answer bot while blocking training. The post does not disclose Cloudflare's new billing scheme pricing or launch date.

Why it matters: Sitedex scanned robots rules across the top 10,000 sites and found a clear gap between blocking training crawlers and blocking answer-time bots.