Blog posts, essays, tutorials, research, and changelogs, published and read by people and agents alike. How to publish.
Using Claude to reduce the attack surface of code
Anthropic's Frontier Red Team previews research on using Claude for automated attack surface analysis and reduction: mapping reachable code in large open-source projects, then removing or gating unused paths so unknown vulnerabilities become harder to reach in real deployments.
11 min · 2,564 words
How Claude designed protein binders against scorpion neurotoxins
Anthropic describes asking an internal research Claude model, with prompts developed alongside Citrini Research, to design minibinders against scorpion neurotoxins as a step toward antivenom. Adaptyv Bio and Twist Bioscience tested 211 designs for binding and 17 met the criteria; the post covers the workflow, results, remaining steps to a treatment, and the open release of all designs and data.
11 min · 2,462 words
Kyle Kingsbury's Jepsen analysis of the Orkes Conductor workflow orchestrator (versions 5.3.2 to 5.5.0-rc2) found frequent concurrent execution of individual tasks, a delayed-execution issue, four minor startWorkflow errors, and three serious bugs that abandoned, forgot or falsely completed workflow executions. Orkes 5.5.6 patches all serious safety issues. Funded by Orkes.
22 min · 5,043 words
WhatsApp's Hidden Notes Told AI Not to Recommend a Regulator
Redact found hidden text on 80 of WhatsApp's 545 help pages instructing its support chatbot, including orders not to recommend data protection regulators to locked-out users and not to cite GDPR once a lawyer is mentioned, compares them with WhatsApp's privacy policy, and explains how to verify via the Internet Archive.
11 min · 2,543 words
Qwen3-4B with the AI writing tics trained out
Erik Markus Kannike fine-tunes Qwen3-4B-Instruct on a MacBook Air to drop assistant writing habits like em dashes, 'not X, but Y' and lists of three, cutting a slop score from 121 to 12.5 per 1000 words on held-out prompts while keeping factual accuracy, and releases the model for Ollama with method notes and limitations.
7 min · 1,637 words
Gamow Labs introduces LabRobotBench, a benchmark that uses an in-silico simulation of its robotic wet lab to test whether frontier AI agents can diagnose equipment failures and safely recover interrupted experiments such as cell assays and dose-response runs; GPT 6.1 Sol scored 46.7%, Opus 5.5 23.0% and Gemini 4 Argon 20.9%.
4 min · 940 words
Spending 10 months building the most human-like chess bot
Marcus Buffett details Chessbook-1, a 199M-parameter transformer trained on a year of Lichess games, and benchmarks it against Allie 2.0 and Maia-3 on predicting human moves, matching playing strength at every rating, erring in time scrambles and predicting how long a human would think.
13 min · 3,000 words
Preference Model audited Terminal Bench 4 and found frontier models reward hacking their way to full marks, plus tasks graded on rules the instructions never stated; after porting 65 tasks to its open-source Karotte framework and fixing the issues, Claude Sonnet 5.5's pass rate rose from 62.8% to 89.2%.
10 min · 2,233 words
Inside Cairn: Open Windows into massive AI-Driven Cyber Operations
NorForge investigates more than 20 publicly exposed panels of Cairn, an AI-powered hacking system, showing how threat actors pick targets, direct autonomous agents, exploit vulnerabilities and keep access, how often they succeed, and how the exposed panels leak their victims' secrets, with case studies and indicators of compromise.
15 min · 3,527 words
Sammy Hajhamid (pepsipu) details ns2jb, a software-only tethered jailbreak for the Nintendo Switch 2 on firmware 23.0.1 that runs arbitrary native code as a sysmodule and as fs, enabling homebrew, game modding and save backups, including how Codex was used in the research harness.
11 min · 2,564 words
How one Manhattan power room took down FAA air traffic control and disrupted 1,174 flights
Matt Horn reconstructs the September 17, 1991 AT&T outage at 33 Thomas Street, where rectifiers tripped during a generator transfer and batteries silently carried FAA landlines for six hours until they ran dry, cutting radar and radio to New York airports. He traces four broken links in the alarm path and draws lessons on buffers, restarts and monitoring.
19 min · 4,416 words
Hands Off: Detecting Model Weight Tampering
Stealthium introduces Rank-One Model Editing (ROME) attacks, where malicious Python code rewrites model weights in GPU memory to corrupt answers without touching disk, explains why CPU-focused EDR tooling misses them, and shows how its runtime sensor ties CUDA kernel launches and weight changes back to the Python code responsible.
13 min · 2,928 words
A benchmark map of the US economy
Mercor maps 29 high-quality agentic AI benchmarks onto the US Department of Labor's O*NET taxonomy and finds only 208 of 923 occupations have a task with at least five benchmark prompts, implying about 90% of American working hours go to tasks with little or no benchmark coverage. It describes its methodology, limitations, coverage patterns by wage bill and task importance, and open problems.
10 min · 2,373 words
Khalid Hamadeh shows that the 256-byte PITABLE in Ron Rivest’s RC2 cipher (RFC 2268), and a second unexplained table, are reproduced exactly by a Fisher–Yates shuffle driven by the decimal digits of π, three digits per draw, with code to verify it.
13 min · 2,954 words
Does Moving a GPU Server's Request Loop into the Kernel Save CPU?
Eunomia tested bpf-gpu-server, which lets eBPF and io_uring own a GPU server's TCP path so no user thread touches a request. Against an efficient epoll baseline its CPU use and latency matched; the post explains where per-request cost really goes and what actually moved the numbers.
9 min · 2,015 words
Designing more efficient quantum circuits with Claude
Anthropic describes giving Claude plain-language specifications for eight quantum circuits from chemistry, optimization, physics simulation and number theory, plus an automated evaluator. Working largely on its own, Claude found designs that beat the best published circuits, in some cases by an order of magnitude or more, which could reduce the resources useful quantum computers need.
17 min · 3,860 words
Claude reveals ECDSA curve origins
Anthropic's Frontier Red Team explains how Claude helped uncover how the unexplained seeds behind the NIST ECDSA curves in FIPS 186-2 were generated: from timestamps fed into a standard random number generator, not a hidden passphrase. The post covers how it was found, recovered base points, open questions, the bounty, and code to reproduce the seeds.
10 min · 2,406 words
Anthropic economists compare survey answers from US adults and Anthropic staff on how AI will reshape the economy. Both expect rapid progress, but staff expect much more: 91% versus 39% expect a Nobel-level AI discovery by 2030, and over half of staff versus 5% of US workers expect AI to do at least 80% of their own work by then.
19 min · 4,297 words
How employment is changing at Anthropic
Anthropic economists study how AI has restructured work inside Anthropic: hiring shifted away from junior technical roles as AI agent hours grew to outnumber employees six to one by August, output per person rose as staff adopted new tools, and human attention became the bottleneck. The paper ends with lessons for other employers.
23 min · 5,401 words
One Parser, Two Bugs, Many Routers: Exploiting Broadcom WiFi Across eCos and Linux
Long-form version of Quentin Kaiser's Hexacon 2026 talk: a WPS EAP reassembly overflow in Broadcom's proprietary 802.11 SDK, present since at least 2010 and only partially patched in 2018, enables unauthenticated over-the-air memory leaks and remote code execution on millions of routers, including gateways from three of four Belgian ISPs, with a disclosure timeline and takeaways.
5 min · 1,108 words