Anthropic reported that its Claude AI system, running 950 agents over roughly 21 hours, sifted through about 200,000 reverse transcriptase samples and identified a previously unseen CRISPR-like enzyme system. Human scientists later confirmed the enzyme was genuinely novel, with MIT and Broad Institute professor Feng Zhang calling the RNA-repeat arrays linked to reverse transcriptases 'genuinely intriguing' and worthy of further study. Anthropic says human involvement was limited to prompting the AI and conducting lab verification.
Anthropic released Claude Sonnet 5.5, replacing Sonnet 5 as its medium-sized model, a week after debuting Opus 5.5. The company says the new model runs over 30% faster and uses fewer tokens while keeping Sonnet 5's pricing of $2 per million input tokens and $10 per million output tokens. Sonnet 5.5 is also the first Sonnet-tier model to ship with cybersecurity safeguards similar to those used in Anthropic's top-tier Opus models.
Anthropic released Claude Sonnet 5.5, the second model in its Claude 5.5 lineup after Opus 5.5. The company says it runs over 30% faster and costs up to 30% less per task than Sonnet 5, while keeping the same per-token pricing, and jumps from 10.3% to 70.6% on the Terminal-Bench 4.0 agentic coding evaluation. A third model, Claude Haiku 5.5, is planned for release in the coming weeks.
A blog post by Simon Späti argues that the real risk from AI-generated code isn't poor quality but that engineers and teams no longer understand why architectural decisions were made. He cites a viral tweet describing a large company where specs, tickets, tests and code are all produced via Claude Code, with staff at every level simply prompting the AI and management pressuring teams to ship faster without reviewing output.
Anthropic's Cat de Jong will join Gamma's Grant Lee and a Clay representative on the AI Stage at TechCrunch Disrupt 2026 for a session on how enterprises deploy AI in practice. De Jong, who works directly with enterprise Claude deployments, will discuss where implementations succeed or stall after the initial rollout.
OpenAI, Anthropic, Google, Microsoft and xAI all allow users to opt out of having their chat conversations used to train future models, a process that takes about a minute per service and doesn't affect personalization or memory features. Meta is the exception: US users cannot opt out of training on Meta AI chats, and since December the company has used those chats for ad targeting without any opt-out option.
Nvidia has moved its OpenShell agentic AI sandbox, first unveiled at GTC in March, into general availability for all users. The tool isolates AI agents' activity at the operating system kernel level while they perform tasks, and Nvidia says it has security collaborations with companies including Anthropic, Cisco, CoreWeave, Dell, Microsoft, Palantir, Salesforce and SAP.
Anthropic released documentation detailing how to prompt its Claude Opus 5.5 model, highlighting behavioral differences from Claude Opus 5 including effort calibration, thinking behavior, and handling of unattended multiagent tasks. The guide notes Opus 5.5 generates output over 30% faster than its predecessor while typically using fewer tokens to complete the same task, and that existing Opus 5 prompts should still work without modification.
A writer describes rebuilding their personal homepage entirely through Claude, without writing or running code or opening a GitHub pull request, using Opus 5.5 in a Claude Project. Roughly 60 parallel Claude sessions handled tasks including a CGI parody video and an original soundtrack, with most orchestration done from a phone rather than a local machine.
An engineer at Recurse Center tried to solve a 3D-printed physical puzzle where corn-kernel-shaped pieces must slide into a grooved cob without gaps or overlaps. After manually attempting it, they asked Claude to write a solving script, which required some back-and-forth to correct piece shapes via computer vision. Rather than implementing recursive backtracking as expected, Claude's script imported Google's OR-Tools CP-SAT constraint solver library to find a solution.
A user named Cobanov used Anthropic's Claude AI to debloat his four-year-old Android TV through developer/ADB access, instructing it to disable unused apps, shorten animations, and log all changes rather than uninstall anything. He also replaced Google TV's ad-heavy home screen with FLauncher, an open-source app launcher, and reported the TV ran smoother than when new, all without rooting the device.
Andrew Boyd, founder of Standard Agents Inc., says his Jev decision model beat Pokémon Red's Elite Four and Champion, reaching the Hall of Fame on September 23, 2026, less than a week after the run began. Jev isn't a chatbot or LLM; it picks from lists of weighted options rather than reading the screen, while Anthropic's Claude Opus 5 monitored the game log and adjusted those options to help it when stuck. A harness changelog logged 474 fixes, including repeated failures like walking into a shut door 53 times and looping a ladder 124 times.
GoKawiil