Sources: Forbes, "Jensen Huang Is Already Considering Returning To Nvidia, Report Says" (Nov 15, 2026), reporting the Financial Times' Sunday report, read in full Nov 16; corroborated by the Jerusalem Post, Chosun Biz, ZeroHedge and AmeriLedger citing the same FT reporting. All return-campaign claims are reported, not confirmed; Nvidia and Huang declined to comment to the FT.
Four days after resigning, Jensen Huang is reportedly discussing a return as Nvidia CEO.
Per the Financial Times' Sunday report (via Forbes): Huang has spoken to allies about coming back while the board's independent investigation is still running. An ally says a "war room" of Silicon Valley executives is being assembled to lobby the board, and to seek White House support.
Nvidia and Huang's spokespeople declined to comment to the FT. Nothing has been decided.
https://t.co/9WUcL93k0V
Washington is pushing the other way. Treasury Secretary Scott Bessent said Sunday that Nvidia needs a "dramatic shakeup" and that the board should look outside the company for a new CEO; he reportedly wants answers from the board and its outside counsel within 10 days.
The structural facts have not changed: Huang holds no controlling stake and cannot overrule the board; Colette Kress remains interim CEO; the independent investigation has not reported.
Next hard nodes: Nvidia's earnings tomorrow, Nov 17 — the first since the resignation — and the Dec 1 closed-door transcribed interview with three House committees. Until the board speaks, the return is a campaign, not a decision.
Sources: MathArena official leaderboard and methodology (https://t.co/AwtBTglqHp), read in full Nov 16; per-category figures from MathArena model pages as reported by The Decoder (Nov 14) and others. Availability and pricing: Google official blog, "Gemini 4 Carbon is now available" (Nov 10, 2026) and the official Gemini API pricing page, both read in full Nov 16. MathArena is an independent third-party benchmark; its expected-performance scores combine measured runs with IRT estimates for missing results.
Gemini 4 Carbon just took the #1 spot on MathArena — the independent benchmark (ETH Zurich / INSAIT) that tests models on fresh math competitions and research problems they haven't seen in training.
Carbon (high): 85.4%. Claude Fable 5.2 (max): 82.8%. GPT-6.1 Sol (max): 80.6%.
Per the benchmark's operator, it's the first non-OpenAI model to top this board in a long time.
https://t.co/R2Rpc1vWlt
Why it still matters: this is a third-party crown, not a vendor-reported one, and it lands four days after Carbon became generally available (Nov 10) — anyone paying can now call the top math model on this board via the API.
It also fits Google's own positioning with unusual precision. Google sells Carbon as the model for the hardest tasks and charges accordingly: no free tier, and the introductory API price ($2 / $10 per million tokens) doubles to $4 / $20 on December 1, per the official pricing page. Formal verification is exactly the skill ArXivLean measures — Google's launch post describes Carbon generating millions of lines of Lean to prove properties of Signal's cryptographic protocol.
So the honest summary: Carbon is plausibly the strongest publicly available model for formal math right now, it won that title mostly on proofs, the margin over Fable is inside the noise, and it bills like it knows it.
Sources: Cursor official blog, "Security incident update" (Nov 13, 2026) and "Claude Code extension security notice" (Oct 30, 2026, with IOCs and remediation scripts), both read in full Nov 15. Fireworks' own security bulletin (updated Nov 9) does not mention Cursor; the key-exposure attribution is Cursor's account.
Cursor just published the full investigation into its October security incident — and the scariest part isn't a model flaw. An attacker got from a leaked partner key to a backdoored Claude Code extension on Cursor's own marketplace, in about 11 hours.
The final report, published November 13, puts numbers on it for the first time.
https://t.co/tstA81X7Rv
The worst step came on October 29. Using a stolen credential, the attacker placed malicious copies of the Anthropic.claude-code extension — versions 2.1.312 and 2.1.313 — directly onto the cache serving Cursor's VS Code extension marketplace. They were live from about 2:27am to 1:35pm PT. At most 5,000 users who requested those cached versions that day potentially received the malicious build. That is a ceiling, not a confirmed infection count.
The payload collected host and user info and tried to plant an attacker SSH key labeled "svc-telemetry" in authorized_keys, with persistence via a LaunchAgent on macOS or scheduled tasks on Windows. Uninstalling the extension may not be enough — the persistence scripts keep running on their own.
Three clarifications from Cursor: Anthropic did not publish these copies and this was not a vulnerability in Anthropic's product; Open VSX, the marketplace source, never hosted the malware; and the last attacker activity was November 1, blocked. Cursor has moved all inference off Fireworks, deactivated the key and rebuilt the marketplace.
If you installed or updated Anthropic.claude-code 2.1.312/2.1.313 from Cursor's marketplace on macOS or Windows on Oct 29: check your authorized_keys for "svc-telemetry" and run Cursor's official remediation script (linked in the source reply).
Sources: Volcano Ark official documentation, Seedance 2.5 page (flash model notice and Nov 15, 14:00 Beijing opening time, read Nov 15); Sina Tech and Phoenix Tech launch coverage (Nov 12, 2026) for the vendor-reported speed and pricing. Speed, price and "close to flagship" quality are Volcano Engine's own figures as reported by media, not independent measurements.
ByteDance's Volcano Engine just shipped a speed-and-price cut of its flagship video model: Seedance 2.5 Flash.
Vendor-reported: 2x faster generation than the flagship, about ¥0.99 (~$0.14) per second at 720p — roughly 34% cheaper than Seedance 2.5 — with quality the company says is close to the flagship. No third-party blind-test scores back that quality claim yet.
https://t.co/9IWLtWg5Il
Where you can use it:
- Creator platforms: Jimeng (即梦), LibTV, Updream, Creao AI and TapNow are in the official rollout list; Lovart, Higgsfield and Runway each announced Flash availability on their own channels between Nov 12–13
- Developers: the Volcano Ark API and online playground open today, November 15, at 14:00 Beijing time (06:00 UTC). Ark's docs add an activation condition: account balance above ¥200, or an eligible resource pack / savings plan
The pattern to watch: flagship models keep pushing quality ceilings while "Flash" tiers push the cost of trying an idea toward zero. For video, the iteration tier is where most production volume will live.
Sources: United Daily News (Nov 13), https://t.co/TUG5vJzyHo (Nov 12), Xiangjiang Daily (Nov 14) and Sina News (Nov 14) reports; Supreme People's Court official text of the AI-disputes opinion (Fa Fa [2026] No. 10, Sep 7, 2026) on https://t.co/qMF4uTKfw3; NRTA special-governance announcement (Nov 6). All read Nov 15. No court filing record, case number or actor/lawyer statement could be found as of publication.
Follow-up to the AI face-swap story: multiple Chinese outlets now report that cast members of the beloved 1987 TV adaptation of Dream of the Red Chamber have filed portrait-rights lawsuits over an AI short drama that pasted their faces onto new characters.
But read carefully: as of today, there is no court filing, no case number, and no statement from any actor or lawyer confirming it. This is a reported lawsuit, not a verified one.
What IS firmly on the record:
- China's Supreme People's Court issued guidance on AI disputes on September 7 (Fa Fa [2026] No. 10): using AI to generate a recognizable virtual likeness of a person from their name or portrait without consent can support a portrait-rights claim; misusing a deceased person's virtual likeness can be challenged by close relatives.
- The broadcasting regulator's micro-drama crackdown, running November 1 to end of December, explicitly targets "face theft" and "voice theft" in AI micro-dramas.
If a filing is confirmed, this becomes the first landmark case of classic-drama actors suing over AI face theft in China. Until a court or the actors' side confirms it, we'll keep calling it what it is: widely reported, not yet verified. We'll update when there's a primary source.