Anthropic / Claude

Reuters reports that Anthropic accuses Alibaba of illicitly extracting Claude model capabilities. The article does not detail the extraction method or which model versions are involved. No response from Alibaba is included, and no technical specifics are provided. I'd hold off until both sides show evidence.

Anthropic publicly accuses Alibaba of illicitly accessing its AI models. The post does not specify whether this refers to model weights or API outputs. Only the headline is available so far—I'll wait for more details before drawing conclusions.

Bloomberg reports Google is about to lose two more senior AI researchers to Anthropic. This is the latest in a string of Google-to-Anthropic moves, though the article does not name the two staffers or specify their roles.

In New York's 12th District Democratic primary, the candidate backed by Anthropic and the one backed by OpenAI fought to a draw. Anthropic's pick Alex Bores didn't win, but OpenAI didn't crush him either. Combined spending hit $27 million; the post doesn't break down how much each side put in. The race was seen as a stress test of both AI companies' political influence, and neither walked away with a decisive win.

At Config 2026, Figma turned its canvas into a workspace for code, motion, 3D, and shaders. Code Layers puts design and production code side by side; Motion brings animation timelines into collaborative editing; Shader uses WebGPU for material effects. But the company admits high inference costs from third-party AI models are squeezing margins, and those models come from providers like Anthropic that are building competing products. Figma's bet is on AI that produces tweakable tools rather than one-shot outputs, plus team-shared prompts and plugins to cut token use. The post doesn't spell out progress on in-house models.

Google added a native computer-use tool to Gemini 3.5 Flash, letting the model take screenshots, click buttons, and fill forms to operate web and desktop UIs. It joins Anthropic and OpenAI in baking screen control directly into a model. The post claims 3.5 Flash beats Claude Sonnet 4 and GPT-5 on the WebVoyager benchmark, but Google didn't release full eval details or reproduction steps—hold for third-party tests. Available now via Gemini API and Google AI Studio; pricing and rate limits aren't disclosed in the post.

NSA cybersecurity analysts had been testing Anthropic's Mythos model and found it could break into nearly all of their classified systems within hours. After the Trump administration imposed export controls on Anthropic this month, Mythos 5 and Fable 5 were pulled back, cutting off NSA's access. Senator Mark Warner cited NSA chief Gen. Joshua Rudd at a hearing, saying Mythos breached classified networks 'not in weeks, but in hours.' The article doesn't say whether NSA has a backup plan or what specifically triggered the export controls.

Anthropic put Claude into Slack as a persistent team member that can work across channels, follow up proactively, and wait days for dependencies. Internally it already writes 65% of product PRs, including most of Claude Tag itself. Beta is limited to Enterprise and Team plans; the post doesn't disclose pricing.

Anthropic's Claude Tag puts a persistent AI teammate in Slack that can proactively chime in and retain channel context. Under the hood, ambient mode is still scheduled triggers hitting an HTTP endpoint, and memory is raw chat logs, not organizational knowledge. The real shift is governance: enterprises are assigning agents independent identities, permissions, budgets, and audit trails. Microsoft's Entra Agent ID treats agents as first-class directory citizens. Pricing is moving from per-seat to consumption-based. One Reddit user burned $41,952 in a month running persistent agents, 99.6% spent replaying context. Continuous learning isn't here yet—Gartner predicts over 40% of agent projects will be abandoned by 2027—but the direction is set: enterprise software's runtime object is shifting from apps to governed agents.

An Anthropic customer is suing the US government after losing access to the Fable model. The article doesn't specify which Anthropic model Fable refers to. The plaintiff argues the government cut off a paid service without due process. Only the lawsuit filing is confirmed so far; case details and the government's response aren't public yet.

Claude Tag lets teams @Claude in Slack channels to delegate tasks. The model breaks down requests, uses connected tools, and works asynchronously. It retains channel context so you don't repeat yourself. Anthropic says 65% of its product team's code is now created by an internal version of Claude Tag, with use cases extending to metrics, support tickets, and bug hunting. It runs on Opus 4.8 and is in beta for Enterprise and Team customers. The post doesn't specify a timeline for expanding beyond Slack.

Anthropic launched Claude Tag, an always-on AI teammate inside Slack. It reads channel messages over time to build institutional knowledge, not just answer one-off questions. The play is clear: turn organizational context into Claude's long-term memory and embed the model into daily workflows. The post doesn't disclose pricing or launch date. Privacy and permission controls are mentioned only briefly—admins can configure them, but details are thin.

David Rosenthal breaks down the subsidy math behind AI platforms. SemiAnalysis found a $200/month subscription can burn $14,000 in OpenAI tokens or $8,000 in Anthropic tokens—a 40–70x subsidy. Ed Zitron obtained OpenAI's 2025 financials: $13.07B revenue against $34B in costs, a $38.5B net loss, with sales and marketing alone eating 44% of revenue. The post argues the 'first one's free' drug-dealer model is hitting a wall as business adoption fails to keep pace with the cash burn.

Anthropic analyzed ~400K real Claude Code sessions and found managers scored highest on verified success—tasks requiring git commits, merged PRs, and passing tests—beating software engineers. The group flagged selection bias: managers chase aha moments, not corner cases. Sakana AI launched Fugu, a multi-model orchestration product hitting SWE Bench Pro 73.7, explicitly marketed as free from US export controls. OpenCode's public dashboard shows 136K DAU with DeepSeek at 53.6% share—domestic Chinese models dominate. On the practical side, the root cause of Gemini 3.5 Flash output truncation was traced to thinking monologues consuming tokens before the response could start; increasing the output limit helps but remains unstable.

SpaceX signed its third GPU deal, leasing GB300s to Reflection AI at $150M/month. Combined with earlier Anthropic ($1.25B/month) and Google ($920M/month) contracts, Jamin Ball estimates SpaceX's GPU revenue at $2.32B/month, annualizing to ~$28B — roughly double CoreWeave's current revenue. Blackwell pricing exceeds $10/hour, which is on the high side. The post doesn't disclose total GPU capacity, utilization, or whether these deals are exclusive.

The author built a blind benchmark from 9 real bugs found by Anthropic's Mythos. Models get the source tree and a target file, no hints. All models underperformed expectations. Gemma 4 MoE led with 4/9 detections but had multiple retries due to crashes; GPT 5.5 Pro burned $100 after only 4 cases. The post doesn't report Mythos's own score on the same setup, nor whether these bugs were found in one shot. Sample is tiny and single-run, so treat rankings as directional.

Sakana AI's Fugu is a trained multi-model orchestrator. It decides which models to call, how to assign roles, and how to verify results—no hand-coded rules. Two ICLR 2026 papers back it: TRINITY trains a sub-20K-parameter coordinator via evolutionary strategy, Conductor trains a 7B orchestrator via RL. Fugu Ultra scores 93.2 on LiveCodeBench, beating Fable 5's 89.8, but trails on SWE-Bench Pro and HLE. Sakana deliberately hides which models are called per request and their raw outputs, calling routing info proprietary. All data is self-reported with no independent third-party replication and no head-to-head against OpenRouter Fusion. The direction holds, but transparency is still missing.

Anthropic is bringing the full Claude Desktop experience to AWS, Google Cloud, and Microsoft Foundry. Enterprises can now deploy Claude Desktop directly inside their own cloud environment without routing through the public internet. The post doesn't spell out pricing or exact launch dates, but confirms this targets Claude Enterprise customers. For teams with strict compliance needs, keeping data inside their own cloud account is a concrete win.

A tenured professor and editor-in-chief argues academia's reward system is broken under AI. Students use two paid models to polish assignments into undetectable, high-scoring work, penalizing honest writers. In research, someone with Consensus and Claude can produce a publishable paper a day, overwhelming peer review. Grant applications face the same volume attack: a five-person team can submit ten proposals per cycle by rotating lead applicants, with AI fixing budget errors and missed citations. The author says institutions are moving too slowly—teaching assessments are being redesigned, but research metrics still treat AI-inflated output as real.

Patrick McCanna inspected Claude Code's local session logs and found that 'thinking blocks' contain only a 600-character signature, not readable reasoning. Anthropic encrypts the actual reasoning into that signature, holds the decryption key server-side, and the API returns a summary. Full thinking output requires an enterprise agreement. The 'extended thinking' you see in the terminal is a post-hoc summary by Fable/Opus, not the raw reasoning that drove the agent's actions. Don't count on this as an audit trail, and the docs are indirect enough that you might miss the caveat without coffee.