Latest news, beside the leaderboard

On November 13, Sierra co-founders Bret Taylor and Clay Bavor announced Personal Agent Protocol (Poppy) v1.0 on the [official blog](https://sierra.ai/blog/releasing-poppy). It is now on for everyone who uses Muse, Meta's AI agent, which looks for Poppy first on every site it visits, and Shopify has enabled it for every merchant on its platform, with no extra work from the merchant. Sierra says 60 more companies have joined as design partners, bringing the total to more than 100, and that Mercury and Turkish Airlines are among those already live. The specification is at [personalagentprotocol.org](https://personalagentprotocol.org/docs/1.0/spec).

In an AWS Security Blog post on November 13, Amazon CISO CJ Moses wrote that from October 3 to November 5, 2026, Amazon's global decoy sensor network MadPot, configured to present as popular LLM chat-completions interfaces, received requests from attack agents running on ARTEX, an open-source agentic penetration testing framework. Because agent frameworks send their system prompts, tool histories and knowledge-graph state to the "model", single requests often carried a campaign's full record. Amazon assesses with moderate confidence that the activity came from disparate Chinese-nexus groups. The largest campaign stole AI model credentials at scale, with stolen keys paying for the inference behind further key hunting; another tried to beat KYC checks at two crypto exchanges with real-time deepfake video and forged ID photos, but the vendor's server-side verdict rejected every recorded deepfake attempt. Amazon stresses that none of the techniques were new and that it cannot verify the outcomes the agents reported.

At 22:37 SGT on November 14, Elon Musk (@elonmusk) quoted a clip of SemiAnalysis founder Dylan Patel and wrote: "Jensen drove supply chain brilliantly at Nvidia, but many of those deals were a handshake with Jensen, not a written contract. Without Jensen, at least some of those critical deals will be lost." The quoted clip, posted by @CEOinterview, has Patel saying Jensen personally talked small suppliers with only a few hundred million dollars of revenue but an important subcomponent into letting Nvidia buy up their supply, and that with him gone, "Google's begging for more capacity from the whole supply chain." Musk's post has about 3.8M views, 56,967 likes, 5,374 reposts and 2,280 replies. This piece covers personal opinions and community discussion: Nvidia has not disclosed how such deals are structured, and neither the "handshake deals" claim nor the "Google begging for capacity" claim has been independently verified.

Podcast host Dwarkesh Patel published a long essay on November 13 subtitled "Why I now expect superintelligence by 2030." He writes that a few months ago he thought superintelligence was at least the better part of a decade away, and now thinks AIs as good as the best humans at AI research are only a couple of years, possibly a couple of months, away. His reasoning: AI has rapidly broken through verifiable domains like math, and AI research is itself largely verifiable. The post has about 820K views and 4,129 likes, quote posts from people like Dan Robinson spread the debate, and X lists it among trending stories. This piece covers one person's views and community discussion, not a verified forecast.

On November 13 the Federal Reserve published its semiannual Financial Stability Report. Of 23 market contacts surveyed by the New York Fed in September and October, 87% cited AI as a risk to financial stability, up from 50% in the spring and the most of any risk. A box in the report says advanced computing debt issuance passed $680 billion by October, about five times all of 2023, and that leases, guarantees and equity stakes from hyperscalers and GPU makers create a "circular financing" dynamic.

xAI (signed as SpaceXSI in its docs) has rewritten the "Output Watermarking" page of its API docs: the invisible text watermark that was due to turn on by default on November 7 is now documented as off by default, and team admins can turn it on per model in the Console for grok-4.7, 4.6, 4.5, 4.3, and the three grok-4.20 variants. The page also says the text watermark uses Google DeepMind's SynthID Text. Audio from the voice APIs (Text to Speech, Speech to Speech, custom voices) is watermarked with AudioSeal by default and can be opted out per model, but stays watermarked when the end user is in the EU, the EEA, or California. On November 14, Elon Musk posted on X that text watermarking with Grok is only on where required by law or requested by the user.

At 03:56 Singapore time on November 14, former Anthropic researcher Jacob Coxon (@hilbertspaess) posted a slide from Anthropic's IPO roadshow on X and said it was a big factor in his decision to leave. His reading: the slide shows pretraining compute efficiency over time on a log-scale y-axis, and the curve is still bending upward, which he calls continued superexponential algorithmic progress, with Mythos just one point on that curve. At the time of writing the post has about 1.19M views, 5,212 likes and 321 replies, and X lists the controversy among its trending stories. Critics point out the chart has no axis labels or numbers. This piece covers one person's interpretation and community reactions, not an official Anthropic conclusion.

At 12:19 Singapore time on November 14, T3 founder and developer YouTuber Theo (@theo) posted on X: "Hot take: Software will get more stable overall thanks to AI." At the time of writing the post has about 109K views, 3,296 likes, 365 replies and 62 quotes, and X lists the debate among its trending stories. Supporters say AI makes writing tests, reading legacy code, tracing changes and fuzzing much cheaper; skeptics worry that AI writes code far faster than humans can review it, so unread code ships and systems fail in stranger ways. This piece collects community opinions, not a proven conclusion.

On November 13, Vahab Mirrokni, Google Fellow and VP at Google Research, shared new results from Antigravity's /teamwork multi-agent system on the Google Research blog and X. Researchers from Google, CUHK, USC and CMU used Gemini 4 Argon and other Gemini models to prove information theory's 2013 Courtade–Kumar conjecture in full generality and formalized it in 50 million lines of Lean 4. /teamwork-LEAP also produced a fully autonomous formalization of the Weil conjectures (about 1.8 million lines of Lean across 28,800 subagent runs). On the engineering side, the rest of the Zircon microkernel was rewritten into about 800,000 lines of safe Rust, and 17 JAXBench TPU kernels got a 4.09x geometric-mean speedup. /teamwork and /boost are available in Antigravity now, and 10,000 academic researchers can apply for six months of free access to the top usage tier.

On November 13, ggml-org published llama.cpp v1.0.0 on GitHub and announced on the Hugging Face blog that both the llama.cpp and ggml libraries are now at v1.0.0. From here on, breaking changes to the public C API in ./include bump the major version, backward-compatible features, backends and model support bump the minor version, and pure fixes bump the patch version. The release removes 35 deprecated functions, promotes the NextN/speculative decoding API into llama.h, adds OCP FP8 KV caches and a MoE expert cache with a configurable CPU share, supports new architectures such as Kimi-K3.1-Pro, hardens llama-server against SSRF, CORS and DNS rebinding issues, and adds an OpenAI-compatible /v1/audio/speech endpoint. Creator Georgi Gerganovs release post has about 4,180 likes.

At 08:34 Singapore time on November 14, Atreides Management managing partner and CIO Gavin Baker (@GavinSBaker) posted on X a letter that he and Altimeter Capital CEO Brad Gerstner sent to NVIDIA's board. Writing as shareholders, they ask the board to run a "slow, measured, and deliberate process" on NVIDIA's future leadership that leaves all options open, including Jensen Huang's return, and to avoid rushing to replace him before the facts are known; they add that Interim CEO Colette Kress and her team can lead capably in the meantime. The post has about 1.35M views, 9,580 likes and 1,160 reposts. Earlier, Elon Musk replied that "a petition is underway to reinstate Jensen and I'm signing it too," a reply with about 207K likes. Baker later said the petition had roughly 20,000 signatures, including about 2,000 NVIDIA employees (his own claim, not independently verified). There is pushback too: Tesla senior staff engineer Yun-Ta Tsai argued that employees who really want Huang back should resign to make the point. This is a community discussion; NVIDIA's board has not publicly responded to the letter or the petition.

LangChain co-founder and CEO Harrison Chase announced on November 13, on the company blog and on X, that LangChain has raised a $300 million Series C at a $3 billion valuation. Existing investor CapitalG led the round, with new investors GIC and DFJ Growth and existing investors IVP and Sapphire Ventures, bringing total funding to $460 million. The money backs LangSmith, its framework-agnostic agent platform, which the company says more than 7,000 organizations use; new LangSmith capabilities will be shown at a release event on November 20.

Microsoft AI announced on November 12 that MAI-Transcribe-2, its in-house speech-to-text model, is generally available in Microsoft Foundry for production use at $0.10 per hour of audio, a limited-time price through the end of 2026. The model entered public preview on September 3 and covers 60 languages in one model, with built-in speaker diarization, word-level timestamps and keyword biasing. Microsoft says it transcribes an hour of audio in about ten seconds of inference and averages 3.4% word error rate on the top 25 FLEURS languages.

Data notebook company Deepnote open-sourced Agent Initiative Benchmark v1.0. It drops an agent into the Slack of PageMantle, a synthetic company with six years of records across 11 services, and checks whether it investigates unprompted, decides whether to speak, and picks the right audience. Across 50 situations and 12 models, GPT-6 Astra scored 8.97, GPT-6.1 Sol 8.65 and Claude Fable 5.2 8.17.

Google Antigravity introduced Dynamic Workflows: type /run-workflow and the main agent writes a Python script that schedules subagents with parallel() and pipeline(), each returning a structured result checked against a JSON Schema. You approve the script before it runs and can save it for reuse. Available on Google AI Pro and Ultra in both the App and the CLI.

ClickHouse released Loom, an open-source (Apache-2.0) agent memory system that keeps memories, a code index and run logs in one ClickHouse database, links memories to code, and flags memories that may be stale when code changes. It supports MCP, an HTTP API and Python/TS SDKs, with Claude Code and Codex plugins. In ClickHouse tests, the mean code-QA score rose from 0.78 to 0.84.

In a Connectionism blog post, Thinking Machines explains how Inkling 2 Small used a native 4-bit (NVFP4) forward pass from the first pretraining step through the final RL rollout, with MXFP8 backward passes using stochastic rounding and quantized communications. The lab calls it the largest model it knows of trained end-to-end with a native 4-bit forward pass, with an 8% wall-clock speedup in pretraining and 20% in RL over full MXFP8 training.

On November 14 Artificial Analysis released Coding Agent Index v2.1, swapping FrontierCode 1.1 for its implementation of Cognition's FrontierCode 2.0 (132 real-world tasks graded on whether a maintainer would merge the change), upgrading Terminal-Bench from 4.0 to 4.1 and keeping SWE-Atlas-QnA-AA. Claude Fable 5.2 (xhigh and high, in Claude Code) and Claude Opus 5.5 (max) share the lead at 58. Because FrontierCode 2.0 is harder, scores fall 5 points on average; only Codex with GPT-6.1 Terra (max) held within 1 point of its prior score.

On November 13 Artificial Analysis said Speechify's Simba 3.5 text to speech model is #1 in its Controlled Voice Arena, statistically level with HeyGen Voice. Every model on this board reads with the same 8 cloned voices (4 US, 4 UK) and users vote blind, so it compares models rather than showcase voices. The leaderboard lists Simba 3.5 at 1192±17 Elo from 1,783 samples, HeyGen Voice at 1190±13 and Qwen-Audio-3.1-TTS-Plus third at 1176; Simba 3.5 costs $6.58 per 1M characters on the API versus $30.00 for HeyGen Voice.

On November 13 Artificial Analysis published the Diarization Index, a benchmark for whether speech to text APIs attribute each word to the person who actually said it. It scores Word Diarization Error Rate (WDER), the share of words given to the wrong speaker or to no speaker, across 97 English conversations of two to eight people, averaging three speaker bands (2, 3-4 and 5-8) with equal weight. NVIDIA's open-weight Nemotron 3.5 Transcribe leads at 3.3%, ahead of Voxtral Small Transcribe 3 at 7.5%, the most accurate priced API; MAI-Transcribe-2 scores 8.8% and is one of the fastest.