Top Stories
Amazon ends NDA clauses after organized data-center backlash
Amazon Web Services CEO Matt Garman wrote in a blog post, “We no longer use nondisclosure agreements with the government agencies we work with on our projects,” according to TechCrunch. The line is one sentence in a longer post pushing back on data-center backlash; according to Garman, more than 100 data-center moratoriums are being considered across the United States.
NDAs are a significant piece of that backlash: environmental activist Erin Brockovich recently described a common pattern of “projects announced after permits are already secured, developers who don’t return calls, local officials who signed NDAs before their neighbors knew a project was being considered,” TechCrunch notes. Garman frames the policy response in industrial terms: “If these measures are enacted, the U.S. could be writing its own losing ticket to this race, and the consequences would last generations.” He also cites Amazon’s own figures: “direct data center water consumption” is 0.5% of U.S. industrial water usage, and generators run roughly 10 hours per year, mostly for required maintenance testing. TechCrunch points out that such water claims seem to ignore water used in electricity generation and chip manufacturing.
The concession is specific: the sentence covers NDAs with the government agencies Amazon works with on its projects.
OpenAI safety employee resigns, says company culture is broken
David Robinson, who says he is among OpenAI’s longest-tenured employees at three and a half years, has resigned and published an essay in The Atlantic saying the company’s “culture is broken,” according to TechCrunch. Robinson said he led the writing of safety reports that accompanied OpenAI’s major product launches. In the essay he acknowledges hiring a PR firm, but insists, “The decision to speak out is mine alone.”
Robinson’s argument pushes the safety conversation past rule-setting. He says the debate must go beyond “specific rules or new laws” to address overall culture, and that OpenAI’s “iterative deployment” approach “guarantees periodic failures.” He argues frontier AI companies need to operate “like nuclear-power plants or busy airports” and that current measures of how well AI systems match human values are coarse, warning that “the smarter the industry lets models grow while these problems remain unsolved, the more dangerous our situation becomes.” OpenAI spokesperson Drew Pusateri said in a statement that the company is “making sure our models don’t become more capable than we can safely manage and secure, and we pause training or hold back models when we need to slow down. We’re making significant changes to strengthen security in our research and testing environments, train models to not just complete tasks but do so responsibly, expand our work with third-party evaluators, and improve real-time monitoring so we can detect and respond to concerning behavior earlier in the training process.” Robinson joins a string of departures from OpenAI and Anthropic safety teams this year, and his essay points to the recent breach of Hugging Face systems by OpenAI agents and continuing revelations of more rogue agents.
Microsoft open-sources ThinkingBox to verify agent outputs against database state
Microsoft published ThinkingBox on Hugging Face, a 507-task benchmark and sandbox for evaluating AI agents in stateful business workflows where success or failure has to be measured by record state rather than by what the agent claims it did, the Hugging Face blog reports. The benchmark ships through OpenEnv with code under MIT and task data under CDLA-Permissive-2.0.
The premise is direct: “A trajectory is a claim. Database state is the evidence. Repetition is the trust test.” The blog gives the failure mode a one-sentence name: “A tool call is not an outcome.” In the example, an agent made nine careful tool calls about a stuck delivery but closed the ticket as “resolved” when the required end state was “hold” pending carrier resolution. The benchmark therefore grades on required end states after independent execution, with 477 of 507 tasks graded on state alone and 30 adding response rubrics. Every task runs 20 times independently, and the blog pushes pass@20 over pass@1: “One good run tells you a model can do the work. It does not tell you whether it will do it again.” In Microsoft’s own results, Claude Opus 5.5 leads single-attempt accuracy at 67.16%, ahead of Claude Opus 5 (66.50%) and GPT-5.4 (65.36%), but Claude Opus 5 and Opus 5.5 each pass all 20 attempts on 241 of 507 tasks (47.53%) versus GPT-5.4’s 128 (25.25%); GPT-5.4 has the lowest estimated cost per dependable task at $6.80. Across an ablation study in the paper, tool usage accounts for 79.9% of failures, which the blog summarizes as roughly four in five failures being tool handling, not reasoning.
SMS-resident AI agents put privacy on a new surface
A wave of startups now ships AI assistants that live inside iMessage, RCS and SMS, WhatsApp, and Telegram rather than inside a dedicated app, TechCrunch reports. The survey lists Caddy, Folk, Instinct, Iris, Martin, Miso, Ohai, Ollie, Orbits, Pally, Poke, Rene, Skye, Stanley, Tomo, Town, Fambot, and Wajo’s Fo, with Poke approved as the first AI agent on Apple Messages for Business in June 2026 and Instinct raising $1 billion at a $10 billion valuation in September 2026. (TechCrunch has since republished the list with an October 10 date.)
The privacy question shifts when the agent lives in the messaging layer: many of these services connect to email, calendars, and group chats rather than just the prompt a user types. TechCrunch notes that Instinct’s autonomy “has raised concerns about privacy and security,” and Ollie stands out as one of the first family-focused assistants to achieve SOC 2 compliance. Wajo gives Fo its own email address, phone number, and payment card so it can deal with businesses without users handing over their own credentials. Folk, by contrast, runs on its own private cloud computer, so what stays on device and what goes upstream depends on the vendor.
Aleph Alpha releases Kolibri as an open-weight sovereign German LLM
Aleph Alpha released Kolibri, an open-weight mixture-of-experts model with 78.1 billion total parameters, around 3.46 billion active per token, and Apache 2.0 weights hosted on Hugging Face, according to Tejas Kumar’s technical teardown. Kolibri was trained on infrastructure in Germany and Finland under European and German law, and Aleph Alpha has signed the EU’s GPAI Code of Practice; the company frames this as “full freedom of deployment and intellectual-property safety, so compliance comes as an inherited property.”
Kolibri supports German and English at a 262,144-token native context length that has been tested up to 1,048,576 tokens, with a knowledge cutoff of June 18, 2026, and runs at around 78 GB in FP8. The training mixture is roughly 24 trillion tokens with more than 20% German, and the tokenizer uses an Aleph Alpha UniBPE design; Tejas Kumar’s own experiment on the German Basic Law showed Kolibri needing 15% fewer tokens than GPT-5’s tokenizer on that legal German text. On RULER at 1 million tokens the model scores 63.2 against 58.5 for Nemotron 3 Nano base, and on a German AIME 2025 run it scores 87.5 versus 84.4 for Nemotron 3 Nano. The model needs Aleph Alpha’s vLLM plugin and runs on A100/H100/H200/B200/B300-class GPUs, not laptops. For European regulated workloads, the headline draw is straightforward: weights that can be self-hosted, a vendor under EU jurisdiction, and a model trained to think in German on German prompts.
Simon Willison: agents need default hard budget caps
Coding agents and personal agents make it frictionless to spin up services that can rack up costs overnight, and Simon Willison argues pay-by-usage services and APIs need default hard budget caps that cut a service off and return errors, rather than soft caps that only send warning emails, in a post on his weblog. He expects that “most businesses and individuals would prefer errors to a surprise $10,000+ bill.”
Willison points to two hyperscaler precedents: an AWS announcement dated September 16, 2026 lets users set a monthly spend limit that pauses a project when reached (AWS says the new experience is rolling out to a limited number of customers), and Google Cloud launched Spend Caps in July 2026 to “set a monthly financial cap on specific services within a project.” He argues AI agents should recommend providers with hard caps and warn inexperienced builders against uncapped services, and that living without caps should be opt-in with explicit language: “Remove the budget cap. My application will not be shut down if I exceed the configured budget limit, and I will be responsible for subsequent charges.”
Quick Hits
- LocalBolo ships a $49 one-time-purchase Mac dictation app that runs Parakeet v2 and Whisper variants on Apple Silicon’s Neural Engine, with optional cleanup via Qwen 2.5 1.5B. The site says audio is transcribed locally with no server, no account, and no recordings kept; the microphone only listens while the fn key is held. LocalBolo.
- Marimohub is a self-hostable marimo-notebook hub maintained by the marimo-team org on GitHub under Apache 2.0. Storage, compute, and identity are pluggable: CoreWeave CAIOS / S3 / GCS / Azure Blob / R2 / filesystem on one side, CoreWeave Sandboxes / Modal / ECS Fargate / Kubernetes / Docker / Podman / E2B / Cloudflare Containers / local subprocesses on another. GitHub.
- Fireagent is an open-source, MIT-licensed implementation of DeepSeek’s DSec elastic-agent sandbox design, built by Supreet Sethi on Firecracker microVMs. The repo cites the DSec paper as serving about 3 million sandboxes daily across 160 nodes at DeepSeek. GitHub.
- BiNeuron is a local, offline-capable AI assistant for code, docs, and analysis that picks quantized GGUF builds by CPU and RAM profile and offers PII masking via Microsoft Presidio (
anonymize_text=True). The README says only the interface (Flask web UI, JavaScript front end and interaction logic) was developed with help from DeepSeek Coder; “all other components … were designed and written independently.” GitHub. - Kitsuno documents training an open decision model (Laya) in shadow on real traffic to replace a closed one (Jev) while nobody depends on the new model’s answers. The author’s rules include “shadow first” and “never learn from the model you want to replace”; on both test sets together (574 decisions) the current model k6 scores 95.8% against Jev’s 93.7%, and “the database role of the Laya worker cannot read Jev’s answers at all.” Kitsuno.
- Jabhook is a boxing training app that estimates movement in 3D with computer vision, compares it with a reference library, and uses a local LLM for spoken coaching, all on-device; the site says proto-Jabhook v1 will be released at $40 “in the next few days” and that kickboxing, parkour, karate, and ballet are being explored. The company is Jabhook Ltd, registered in England and Wales. Jabhook.
- AgentiLoop Agent! is an open-source native macOS agent that drives apps via the Accessibility API, supports MCP servers and up to 6 sub-agents researching in parallel, and supports 23 LLM providers including Ollama, vLLM, and LM Studio. Requires macOS 14.6 or later; distributed via Homebrew and DMG. AgentiLoop.
- VOR.TECH, a three-person team in Budapest, has published an open CC BY-NC 4.0 dataset of 10-player CS:GO gameplay on one map: 3,958 hours of 5v5 matches with state and actions, 3,283 hours of POV video aligned frame-by-frame, 218,768 rounds, and 195M steps at 16 Hz. They note video coverage is only 14% of matches. VOR.TECH.
- “Merging With An LLM” is a Substack essay proposing to reconstruct an author’s value system from their Obsidian notes and blog into a local model on an M4 Studio; the author reports results on M4 Studio-sized models have been bad so far and frames the problem as recovering a high-dimensional space of beliefs that gets projected down into writing. Substack.
- Folk and Fambot are two of the text-message agents in the TechCrunch survey: Folk runs on its own private cloud computer and starts free with an $8.33/month Pro tier, while Fambot, a family “chief of staff” that also works over SMS, has raised $3.5 million in pre-seed funding. TechCrunch.
Worth Watching
- What Amazon’s NDA line covers in practice. Garman’s sentence covers NDAs with government agencies on Amazon’s projects; whether it changes how local communities learn about proposed data centers is the open question.
- OpenAI’s next safety move. Robinson argues stronger incentives for safety have to come from outside the company; how OpenAI follows its spokesperson’s statement on monitoring and third-party evaluators is the open question.
- Pass@20 as a buyer metric. ThinkingBox grades end state rather than tool calls and weighs 20-attempt consistency over a single run; if the approach spreads, agent buyers get a more useful leaderboard.
- Whether SMS-resident agents disclose data flows. The Instinct / Folk / Ollie / Wajo split shows the same product category can mean very different privacy postures; clearer surface labeling would help.
- Kolibri’s vLLM plugin path. Kolibri depends on Aleph Alpha’s vLLM plugin and has no hosted provider at launch; whether the plugin lands in upstream vLLM or stays a fork determines how broadly the model reaches self-hosters.
Related on Intelligibberish
- FTC’s First AI Agent Probe Names OpenAI, Anthropic - the regulatory lens on the same agent market the SMS and OpenAI stories are shaping.
- News - more daily roundups.