Key Highlights

  • The copyright question is settled less than the headlines suggest. TechCrunch unpacks why Judge Alsup’s $1.5B Anthropic ruling was actually a win for training-on-copyrighted-text — the penalty targeted how the books were acquired (shadow libraries), not the training itself.
  • Flock Safety’s CEO asks for “compromise” after The Washington Post documented 46 cases of officers misusing its surveillance network, including to stalk ex-partners. Opposition is now bipartisan.
  • A stealth frontier model called Ox Alpha appeared on OpenRouter with a 1M-token context window and free access — and nobody will say who built it. Both TechCrunch and The Rundown are chasing the same trail toward Chinese labs.
  • “Coding is solved, engineering isn’t.” Theo’s 34-minute breakdown of the Boris/Matt PCO argument lands on a concrete thesis: agents write fine code, but our codebases are too hard to verify — and that’s now our bug, not the model’s.
  • GitHub Trending is dominated by agent tooling today — openai/codex, NousResearch/hermes-agent, and anthropics/claude-plugins-community are all climbing fast.

Analysis & Opinion

Is it legal to train AI models on copyrighted books? It’s complicated — TechCrunch

The headline number everyone remembers — Anthropic’s $1.5 billion copyright settlement — obscures what Judge William Alsup actually ruled. He found that using copyrighted material to train a model was lawful; the penalty was for sourcing books from illegal shadow libraries rather than for the training methodology. IP attorney Cathy Gellis frames the distinction sharply: “Copyright law hinges on copying, but it doesn’t hinge on using the work or experiencing the work, consuming the work, reading the work.” Alsup went further, comparing how a language model absorbs text to how a writer studies literature for inspiration. The disputes turn on fair use — the doctrine permitting copyrighted material for transformative purposes like criticism, parody, and education — which means the fight ahead is less about whether training is legal and more about how the training corpus was obtained. Authors whose work was ingested without permission may find that acquisition provenance, not consent, is the only lever they have.

Flock CEO calls for ‘compromise’ as surveillance company faces growing backlash — TechCrunch

Flock Safety’s license-plate-reader network is facing the kind of backlash that arrives when infrastructure outruns oversight. The Washington Post documented 46 instances of law enforcement officers allegedly using Flock’s system for unauthorized surveillance — including tracking former romantic partners. CEO Garrett Langley’s response threads a needle that may not hold: he expressed remorse but argued, “I don’t think that Flock created police abuse. I think we’re the first company to ever shine a light on it.” The political opposition is notably bipartisan — Democratic candidates have attacked the sheer density of camera deployment while Republican legislators introduced bills restricting federal purchases of facial-recognition and plate-reader systems. Flock has made defensive concessions: default data retention cut from 30 days to 7, and case codes now mandatory for access. The deeper question the piece raises is whether retention limits are a meaningful control when the abuse pattern is authorized users making unauthorized queries.

Who’s behind the new ‘stealth model’ Ox Alpha? — TechCrunch

An anonymous reasoning model billed for “coding, sustained agentic work, and production workload” landed on OpenRouter on Thursday, and the guessing game has been running ever since. Stripe CEO Patrick Collison called it “very impressive”; AI analyst Andrew Curran initially pointed at GLM models from Chinese firm Z.ai before the evidence muddied. Wccftech floated an unreleased Microsoft MAI variant as an alternative theory, and Reddit is split between dismissing the Chinese-origin theory entirely and treating it as obvious.

A mystery challenger at the AI frontier — The Rundown

The Rundown adds benchmark texture to the same Ox Alpha story: early claims of 80% on DeepSWE deflated to 63% under more careful evaluation — roughly Fable 5 territory, but achieved with significantly fewer tokens per task. The model ships a 1M-token context window with multimodal support, and OpenRouter’s “near-unlimited” free promotional access during launch week has driven heavy developer usage. Naming conventions and response patterns keep pointing back toward Zhipu AI (GLM-5.3 Flash or GLM-6), consistent with a pattern where recent anonymous OpenRouter launches have come from Chinese labs.

New Products & Tools

GitHub Trending — GitHub

Today’s trending list is almost entirely agent infrastructure. openai/codex (Rust, ~117k stars, +1,990 today) leads on daily velocity, followed by AprilNEA/OpenLogi (Rust, +1,102), NousResearch/hermes-agent (Python, +899), and Alishahryar1/free-claude-code (Python, +889). Also climbing: multica-ai/andrej-karpathy-skills, anthropics/claude-plugins-community (+490 on a small 1.2k base), makeplane/plane, apache/maka, MadsLorentzen/ai-job-search, and PostHog/posthog.

Linkdaze’s smart calendar is built to run a household, not just track a schedule — TechCrunch

A touchscreen family hub that syncs Google, iCloud, Outlook, Yahoo, and Cozi calendars with per-person color coding, plus chores, meal planning, and shopping lists. Its most distinctive feature is an AI meal planner with “Snap-to-Sync” — photograph a paper recipe or a school lunch menu and it converts to a digital meal plan with a matching shopping list.

Research

AI Chip Architectures — Lobsters

A survey tracing the “Cambrian explosion of novel computer architectures” that Hennessy and Patterson predicted in their 2018 Turing Lecture, when single-threaded CPU gains had collapsed from 52% annually in the 1980s to just 3%. The piece maps which domain-specific designs actually reached meaningful deployment — NVIDIA and AMD GPUs, systolic-array accelerators like TPUs and Trainium, the Cerebras Wafer-Scale Engine, and Groq’s LPU — with NVIDIA still holding clear market leadership.

Interviews & Conversations

He’s right. — Theo - t3․gg (33:48)

Transcript-based summary. Theo referees the argument between Boris (Anthropic) declaring “coding is solved, bugs are not yet solved” and Matt PCO calling that “a vanilla ice cream of VC funding stirred with a little pinch of turd” — and concludes both are right, because they’re using “coding” to mean different things. His actual thesis is more useful than the referee call: the failure mode in agent-written code isn’t bad code, it’s missing verification. He walks through the Claude Code desktop bug that started the fight — truncated update text that no human or agent ever looked at in a running app — and argues that reading a diff can no longer catch this class of bug, because if a bug were visible in the code, the agent would have caught it too. The structural claim: “If it is too hard for you to spin up and test, it is way too hard for your agents to do the same.” He points at Anthropic as the cautionary example — Claude Code Desktop is an Electron app requiring a full graphical VM per test thread, which is why it improves so much slower than Claude Code Web, where an agent can just open eight tabs. He borrows John Ousterhout’s tactical-vs-strategic framing from Matt’s thread: agents are “tactical tornadoes,” excellent at execution and indifferent to long-term codebase health, so the strategic work (architecture, pipeline design, deciding what’s worth building) stays human — and the concrete homework is rebuilding repos so agents can verify their own output before a human ever sees the PR.

References

  1. “Is it legal to train AI models on copyrighted books? It’s complicated,” TechCrunch, 2026-08-23 [blog]
  2. “Flock CEO calls for ‘compromise’ as surveillance company faces growing backlash,” TechCrunch, 2026-08-23 [blog]
  3. “Who’s behind the new ‘stealth model’ Ox Alpha?,” TechCrunch, 2026-08-23 [blog]
  4. “A mystery challenger at the AI frontier,” The Rundown, 2026-08-24 [blog]
  5. “GitHub Trending,” GitHub, 2026-08-24 [blog]
  6. “Linkdaze’s smart calendar is built to run a household, not just track a schedule,” TechCrunch, 2026-08-23 [blog]
  7. “AI Chip Architectures,” jepeake.com via Lobsters, 2026-08-23 [blog]
  8. Theo - t3․gg, “He’s right.,” YouTube, 2026-08-24 [video]