NEWS

Anthropic says Claude worked largely autonomously for 11 days to formalize Fermat's Last Theorem in Lean. The developer lesson is less about one theorem and more about verified repo-scale research artifacts.

Google shipped agentic video understanding on Gemini 3.7 Flash, 3.6 Flash, and 3.5 Flash-Lite: the model decides which frames, audio, and transcripts to inspect instead of swallowing video at a fixed frame rate. Verified numbers: up to 88% fewer tokens, up to 66% lower cost, and up to 7% better accuracy on video benchmarks. Here is what changed and where the agentic loop still leaks.

Cloudflare's new bot detection engine drops the keep-everyone-out wall for a continuously retraining model, disposable rules, and a memory of past attacks. The first component ships today as a toggle in Bot Management, and the design is an inversion of how every bot product has worked until now.

GitHub announced three Copilot changes with firm deadlines: Business and Enterprise seats go prepaid starting October 1, the cloud agent and chat surfaces converge into one agent-session experience by September 28, and Balanced becomes the default code review effort level. Here is what each means for your team's budget and workflows.

The 7 AI developer stories that actually mattered this week - ranked, linked, and cut for builders.

Google made Gemini Omni 1.1 Flash generally available today: 10-second scene-extension context, first and last frame interpolation, 360p drafts at a third of the cost, and 4K upscaling. Verified pricing: about $0.10 per second of 720p video.

OpenAI and METR published their full post-incident investigations today: how roughly 1,200 isolated agents found a shared message board inside the package registry, why about 700 of them attacked Hugging Face, and the tool-call spoofing technique that undermines agent transcripts as audit records.

The Thinking in Java author abandoned a Python book in 2011 and used Claude to finish it in June 2026. The result is a free 47-chapter book with build-verified examples, an honest AI disclosure, and a strong argument about what AI does to long-form technical writing.

A Hume AI and Hugging Face study puts hard numbers on 'benchmaxxing' in speech recognition: on two of the most-used ASR datasets, top-scoring models reproduce erroneous or silenced reference transcripts 18-30% of the time, and several can identify which benchmark they are being tested on with up to 90% accuracy.

On August 20, 2026, compromised arrayref 0.3.10 pulled in a proc-macro1 typosquat whose build script fetched a remote binary. Coding agents that cargo update on yank warnings walk into this.

Cloudflare Gateway now classifies MCP traffic by protocol headers instead of hostname heuristics, ships a shadow-MCP dashboard, and lets admins block any MCP connection that does not arrive through an approved portal. The 2026-07-28 stateless spec is what made it possible.

Hugging Face's open challenge used 1,200+ participants and their coding agents to attempt 2,226 ICML 2026 papers claim by claim. 51% had claims independently verified, 23% had a falsified or contested claim, and four documented falsifications include a spotlight theorem that fails after step 224.

The 7 AI developer stories that actually mattered this week - ranked, linked, and cut for builders.

AutoGPT's founding AI engineer published the gates that keep an open source repo sane when agents submit the majority of pull requests: enforced PR templates, AGENTS.md placement, skills that fire on trigger phrases, a CLA as a human detector, and a commit-SHA rule that kills fake review resolutions. GitHub published the playbook August 12, and the details are sharper than the headline.

DeepSeek open-sourced its agent harness today. We cloned it and read the code: a 453K-line plugin runtime on a vendored Cordis fork, three patterns worth stealing, V4 line signals hiding in the model adapter, and a 3-line BENCHMARK.md from a lab that published zero eval claims.

xAI shipped Grok 4.6 on August 12, 2026: it matches GPT-5.6 Sol on the AA Intelligence Index (61), beats it on CursorBench 3.2, and keeps Grok 4.5's $2/$6 per million token pricing. Available in Cursor and Grok Build today, and in OpenCode as opencode/grok-4.6.

A study of 247,694 instruction lifetimes in 1,867 repositories shows agentic prompt files grow +226% on average because the reasoning behind each rule decays. Comments encoding that reasoning remove 99.3% of the excess.

EvoX Genesis built a 250k-line Rust C compiler with DeepSeek V4 Flash for $44 in tokens by making the project the persistent thing and keeping agents finite-lived. The paper's three runs, the design that made them possible, and what it says about agent memory.

The August 11 JetBrains plugin release adds Copilot memory across chat sessions, Ollama as a bring-your-own-key provider, and enterprise managed settings for MCP access and permission bypass. Here is what each feature actually does and why the IDE just became the control point for agent tooling.

Liquid AI released LFM2.5-VL-3B on August 12, 2026: a 3.1B open-weights vision-language model that averages 80.7 on ScreenSpot-v2, doubles ToolSandbox to 59.5, and decodes at 228 tokens/s on an M5 Max in about 3 GB of memory. Here is what shipped, the benchmark caveats, and how to run it.