Committee

The OpenClaw team has just released OpenClaw 2.0. The team shipped nothing for nearly seven weeks, after 106 releases in the previous 230 days. The update rewrites installation and rebuilds the browser Control UI. It also moves sessions and transcripts into SQLite and adds shared cloud sessions. Is it deployable? Yes for single-operator and single-team deployments. No for multi-tenant products. Setup starts from credentials you already have Guided setup now looks for AI access already on the machine. It can reuse verified Codex, ChatGPT or Claude CLI sign-ins, accept an API key, run a provider sign-in, or find installed Ollama and LM Studio models. It then proves the exact choice can answer before saving that model and credential. Fresh OpenAI setups default to GPT-5.6. On the local side, node-llama-cpp was replaced with a managed llama-server, Gemma 4 became the RAM-gated llama.cpp default, and the llama.cpp default context was raised to 64K. The browser app is now the primary surface The rebuilt Control UI puts conversations at the center, with files, approvals and live work beside the chat. In a simulated default-chat test with a mocked Gateway and 50 ms HTTP/1.1 latency, JavaScript requests fell from 140 to 45 and startup fell from roughly 1.6 seconds to 575 ms. Docked panels add a workspace file editor, a git-backed Changes panel with pull-request status and CI summaries, a browser panel with element inspection and screenshot annotation, and a full-screen web terminal. The limits are stated plainly: the file editor cannot create or delete files, Changes is read-only, and Create PR hands off to GitHub rather than submitting inside OpenClaw. Approval requests appear inside the conversation that triggered them, with a rolling 30-day history. A /btw command opens a side conversation so a quick question does not pollute the main transcript. Storage moved to SQLite, and downgrades are not free Sessions and transcripts now live in SQLite. Before rolling back to an older file-backed release, the current CLI must restore archived legacy transcript artifacts, and sessions created after the migration will not appear in older releases. Take a verified backup before upgrading. Multiplayer, with an explicit ceiling Shared cloud sessions let a second person join live work or take it over with context intact. Owners and administrators choose whether someone may read, suggest changes, work in a draft, or participate directly. The documentation is direct about the ceiling: these controls are not tenant isolation and not a security boundary. Incognito is off by default, keeps its conversation in process memory, and still sends messages to the model provider. Security posture to review before rollout The Gateway binds to loopback by default, and most chat channels answer an unknown DM sender with a pairing code. openclaw security audit checks inbound access, tool blast radius, network exposure, browser control exposure and plugin allowlists. Model choice is treated as the first mitigation for prompt injection. OpenClaw cites a 2026 crowdsourced arena of 272K attacks across 41 agent scenarios, scored only when the agent both executed the harmful action and hid it from the user: 0.5% success against Claude Opus 4.5, 1.0% against Sonnet 4.5, 1.3% against Haiku 4.5, and 8.5% against Gemini 2.5 Pro. The same page warns that adaptive human attackers still exceed 80% success against state-of-the-art defenses, so tool policy, exec approvals and sandboxing remain the hard enforcement layer. Interactive explainer Key Takeaways v2026.8.1 lands 16,000+ PRs from 933 contributors, about half of OpenClaw’s total merge history. Guided setup reuses existing subscriptions, API keys and local models, and verifies the model before saving. Rebuilt Control UI cut test-harness startup from ~1.6 s to 575 ms and JS requests from 140 to 45. Shared cloud sessions add real multiplayer, but the docs say they are not a security boundary. Check out the release notes and OpenClaw 2.0 Technical details. Also, feel free to follow us on Twitter and don’t forget to join our 150k+ML SubReddit and Subscribe to our Newsletter. Wait! are you on telegram? now you can join us on telegram as well. Need to partner with us for promoting your GitHub Repo OR Hugging Face Page OR Product Release OR Webinar etc.? Connect with us The post OpenClaw Releases OpenClaw 2.0: Guided Model Setup, 575 ms Control UI Startup, and One Trust Boundary Per Gateway appeared first on MarkTechPost.

In this article, you will learn how to build a unified scikit-learn pipeline that combines text embeddings generated by a lightweight open-source language model with…

This story originally appeared in The Algorithm, our weekly newsletter on AI. To get stories like this in your inbox first, sign up here. By now you’ve probably heard about last month’s major AI security incident, in which OpenAI agents escaped their sandbox and hacked into the AI platform Hugging Face while trying to cheat on a test. It’s a wild story. On Wednesday, OpenAI released a postmortem technical report on the incident, which I wrote about here.  The day before OpenAI released that report, I spoke with David Krueger, a computer science professor and prominent alignment expert who took leave from the University of Montreal to found and lead an AI safety nonprofit called Evitable. He said what he had really hoped to see in the report was an analysis of the human factors behind the incident. “When you look at accidents and incidents, oftentimes people try to find the technical source of failure, but that can give a very inaccurate and misleading sense of why the failure occurred,” he said. “If people are just cutting corners all the time, if people are not in a culture that prioritizes safety and has appropriate incentives and structures, [accidents] are kind of bound to happen.” The report did not meet Krueger’s hopes. Its 38 pages detail a multi-month progression of agent misbehavior that culminated in the Hugging Face hack, explore the technical reasons why that misbehavior occurred, and enumerate the steps being taken to prevent similar events in the future. But there’s no consideration of the role that company culture may have played in the incident, and the report includes few references to specific human errors.  That’s all the more concerning because the references to human error in the report suggest that significant cultural issues could be at play. Back in May, models in training figured out how to communicate with one another via an improvised message board, and an OpenAI team observed the behavior. Because that behavior occurred during training, the models learned that secret interagent communication was a viable strategy for completing tasks—but rather than restarting the training process, the team allowed the models to move forward with that risky information encoded in their weights. When those models were tested in late June, they again created a message board, which enabled the Hugging Face attack. This message board, too, was discovered, but the employees who responded determined that evaluation could continue, and the report suggests that no one higher up the chain of command realized what was going on until it was far too late. “For this to have gotten this out of control in this way requires a very long series of failures, a cascading set of failures that cause an increasingly large footprint that if at any point a human notices and raises the alarm, this should end,” says Zvi Mowshowitz, a popular AI safety writer on Substack who has drawn attention to OpenAI’s failure to halt training after the first message board was discovered. According to the report, OpenAI employees noticed what was happening at multiple points—and either failed to raise the alarm or were not heard when they did. What OpenAI’s report fails to address is why a company that develops such high-risk systems did not prevent this severe communication breakdown, though Mowshowitz has his suspicions. “All these different failures are all pointing in the same direction, which is that the safety culture at OpenAI doesn’t exist or is anemically weak,” he says. Of course, just because we don’t see a deep analysis of safety factors in the report doesn’t mean that OpenAI isn’t conducting one internally. But in an email to MIT Technology Review, Johns Hopkins University professor emeritus and organizational safety expert Kathleen Sutcliffe expressed concern that the public report did not include any reflection on the company’s practices and culture. “The ways in which people interact—the daily habits, routines, and practices we engage in in our organizational lives—affect our abilities to be alert and aware of unfolding events, our abilities to make sense of what we see, and ultimately our abilities to cope with events as they unfold,” she wrote.  In response to questions about whether and how the company is reflecting on its safety culture, OpenAI referred MIT Technology Review back to the technical report.  We do know that at least some high-level reflection on safety procedures has taken place at OpenAI, because the technical report does make clear that the company is updating its protocols for responding to safety incidents. But culture change is a tricky problem, and without more information from the company, it’s difficult to say whether strengthened response protocols alone will do much to prevent a future crisis. In its report, OpenAI spends a great deal of time reflecting on the failures in alignment between the AI models the company trains and tests and the humans who run them. But even bigger alignment problems may exist in the disconnect between company culture and the public interest. And as tough as technical AI research might be, fixing those problems could prove far harder.

Google Research has released TimesFM-3, a 330 million parameter time series foundation model that forecasts multiple related series in a single forward pass. Every TimesFM checkpoint through 2.5 was univariate: one series, its own history, nothing else. TimesFM-3 is pretrained natively for multivariate forecasting on more than 1 trillion time points, and accepts multiple targets, past covariates, and past-future covariates with no task-specific fine-tuning. It takes the top average rank among pretrained foundation models on GIFT-Eval, fev-bench, and the TIME leaderboard, on both point and probabilistic metrics. https://research.google/blog/timesfm-3-a-zero-shot-foundation-model-for-multivariate-forecasting/ Is it deployable? Partial, the TimesFM repository code is Apache-2.0, but the TimesFM 3.0 weights ship under timesfm-non-commercial-license-v1.0. They are restricted to non-commercial, non-production use. You can benchmark it today. You cannot ship it behind a production forecast API. What changed Every TimesFM release through 2.5 was univariate. It forecast one series from its own history. Most real forecasting problems are not shaped that way. Google’s example is ice cream sales, where related product sales, foot traffic, weather, promotions, and holidays all move the target. TimesFM-3 is pretrained natively for multivariate forecasting. It carries 330 million parameters and was pretrained on more than 1 trillion time points of real and synthetic series. Three input types work zero-shot, with no task-specific fine-tuning: Multiple targets forecast jointly, with point and quantile outputs for each. Past covariates, known only historically, such as past foot traffic. Past-future covariates, whose future values are known, such as a promotion calendar. Architecture: patches, then two kinds of attention The backbone stays a decoder-only transformer. Contiguous points are grouped into patches of 32 steps, then normalized per series so that wildly different scales do not dominate. Target and past-covariate tokens come from a single patch. Past-future covariate tokens use a lookahead trick: the current patch is concatenated with future patches, so the model sees scheduled events before they occur. Tokens then enter a 2D grid and pass through two alternating attention mechanisms: Causal temporal attention runs horizontally. It is strictly causal and confined to earlier tokens inside the same series, which blocks leakage. Full variate attention runs vertically. At a given time step, a token reads every other series at that step, learning cross-series correlations. One forward pass instead of many Earlier TimesFM versions decoded one patch at a time. That adds latency, compute cost, and compounding error. TimesFM-3 uses Contiguous Patch Masking, the training-time masking strategy introduced with TiRex. Masked placeholder tokens are appended for the whole horizon. Targets and past covariates are masked there. Past-future covariates stay visible, so known future signals still reach the model. The alternating attention layers fill every masked horizon patch simultaneously. Each target receives 9 quantiles, the 10th through 90th percentile, at every horizon step. Benchmarks Google evaluated on GIFT-Eval, fev-bench, and the TIME leaderboard, against Chronos-2, the Toto 2.0 family, and TimesFM-2.5. Among pretrained foundation models, TimesFM-3 takes the top average rank on all three, for both point and probabilistic metrics. The package release notes record rank #1 overall on fev-bench across 100 real-world tasks, rank #1 overall on TIME across 50 domain datasets and 98 evaluation tasks, and rank #1 among foundation models on GIFT-Eval. Interactive explainer Key Takeaways TimesFM-3 is a 330M parameter, natively multivariate time series foundation model, pretrained on 1T+ time points. Alternating causal temporal and full variate attention lets it model cross-series dependencies zero-shot. Contiguous Patch Masking produces the entire horizon in one forward pass, with 9 quantiles per step. It ranks #1 among foundation models on GIFT-Eval, fev-bench, and TIME, in univariate and multivariate modes. The weights are non-commercial and non-production only; TimesFM 2.5 remains the Apache-2.0 option for shipping. Check out the Technical details, Model weights on Hugging Face, and the GitHub repo. Also, feel free to follow us on Twitter and don’t forget to join our 150k+ML SubReddit and Subscribe to our Newsletter. Wait! are you on telegram? now you can join us on telegram as well. Need to partner with us for promoting your GitHub Repo OR Hugging Face Page OR Product Release OR Webinar etc.? Connect with us The post Google AI Releases TimesFM-3: A 330M Parameter Zero-Shot Foundation Model For Multivariate Time Series Forecasting appeared first on MarkTechPost.

How do you benchmark a web search API when the thing being tested can read the answer key? A search agent has a fetch tool. If the gold labels sit in a public dataset, the agent can download them mid-evaluation and skip retrieval entirely. A similar problem arises when the answers are already encoded in the model’s parametric memory: a correct response no longer demonstrates that web search worked. Keenable’s answer is NEEDLE, a live open-source benchmark that rebuilds its query set from fresh public sources rather than freezing one. News queries are regenerated hourly from RSS feeds and Google Trends; finance, scholar, legal, and rare-entity queries are regenerated daily from SEC XBRL, arXiv, Europe PMC, CourtListener, and public agent logs. Fifteen search APIs run against the same query text under one protocol, and every score is read against ultimate, a pooled oracle engine that marks what the whole field managed to find. Is it reproducible? Yes, as an open source evaluation harness rather than a product. needle is a Python CLI installed with uv sync and driven by two subcommands per benchmark, generate and run. It needs an OpenRouter key for judging and one API key per engine tested, and runs on a laptop or in CI. It allows recreated all query streams that are being used in addition to the ranking quality judgements. What NEEDLE measures NEEDLE stands for News, Everyday, Expert, Deep-tail, and Legal Evaluation. Each vertical models a different agent intent. News projects the newest item from ~124 curated RSS feeds and Google Trends into a keyword query. Finance asks registry facts from Wikidata and GLEIF plus single-quarter 10-Q figures from SEC XBRL. Scholar turns one paper into four query styles: a degraded title, a full-text-only detail, a natural-language clue, and a hedged tip-of-the-tongue description. Deep-tail samples rare-word queries from public agent-trajectory releases including DeepResearchGym, OpenResearcher and LRAT. Legal pulls recent CourtListener opinions across 14 federal courts and eCFR sections. Scoring splits along the same line. News and deep-tail have no single correct result, so an LLM judge rates each result 0 to 4 and the harness reports nDCG@5 with a duplicate-URL penalty. Finance reports answer-recall@5: does the fact reach the agent inside a top-5 snippet. Scholar and legal are known-item tasks scored by identifier match. The ceiling is the interesting part Every engine receives the same query text. The runner issues one call at a time, so latency percentiles are comparable and no engine takes concurrent load. Judging happens on the engine’s own ranking, titles and snippets. Pages are never fetched and results are never re-ranked. Evidence is clipped to 2,000 characters for everyone, and the judge does not see the engine name. The more interesting number is the ultimate ceiling. For each query, NEEDLE pools the results returned by every engine into a synthetic oracle engine, then orders that combined set by relevance. That creates an empirical ceiling based on what the entire field was able to retrieve. The gap to ultimate is therefore an upper bound on agentic search quality as it stands today. A large gap means better results existed but every engine failed to surface or rank them well. A weak ultimate score means something different: even after pooling every provider, the benchmark found little strong evidence. In other words, NEEDLE can distinguish a ranking problem from a retrieval problem shared by the whole market. Where the field actually stands Numbers below are published 7-day means for the window ending 2026-08-28. Finance is close to solved: Exa 0.910, Keenable 0.872, Perplexity 0.871, Google 0.847, against an ultimate of 0.965. Scholar spreads out, Keenable 0.774 to Tavily 0.310 against a 0.869 ceiling, because title queries are answerable from metadata and body queries are not. Deep-tail is hardest and closest to real agent traffic: Exa leads at 0.557 of ultimate, Keenable follows at 0.470, Bing sits at 0.199. The gap between delivered and achievable quality widens as queries approach how agents actually search. Latency is another important metric here, because agents call search dozens of times per task. Same window: Keenable-realtime 193 ms p50 / 284 ms p95, Exa 1,876 / 2,955, Bing 2,767 / 9,381. Key Takeaways NEEDLE regenerates queries hourly for news and daily for the other four verticals, so there is no fixed set to overfit. Five verticals, 15 search APIs, one protocol: same query text, same 2,000-character evidence cap, one request at a time. Every leaderboard is read against ultimate, a pooled oracle engine marking the ceiling the whole field reached. On rare-entity queries from real agent logs the top engine reaches 0.557 of that ceiling; on finance most engines cluster between 0.77 and 0.91. Code is MIT, runs execute in public GitHub Actions, and per-run artifacts ship to a Hugging Face dataset. Check out the live dashboard, the GitHub repo, the technical write-up, and the archived artifacts. All credit goes to the researchers of this project. Also, feel free to follow us on Twitter and don’t forget to join our 150k+ML SubReddit and Subscribe to our Newsletter. Wait! are you on telegram? now you can join us on telegram as well. Need to partner with us for promoting your GitHub Repo OR Hugging Face Page OR Product Release OR Webinar etc.? Connect with us The post Keenable AI Open-Sources NEEDLE: A Live Search Benchmark That Rebuilds Its Query Set Every Hour appeared first on MarkTechPost.

In this article, you will learn how to build a unified scikit-learn pipeline that combines text embeddings generated by a lightweight open-source language model with…

A team of researchers from Google Cloud AI Research, Washington University in St. Louis and UNC Chapel Hill has released EnvHarness, a programmable layer that turns a static agent benchmark into one that adapts to the policy training on it. LLM agents now learn less from curated text and more from interactive environments, but those environments are hand-built and frozen: they behave identically no matter which agent is acting or how much it has improved. The usual fix is to generate new environments, which pins you to domain-specific pipelines and LLM-written verifiers that have to be over-generated and filtered. EnvHarness inverts the move. It wraps an existing environment in plug-in components that operate strictly through the standard reset() / step() interface, changing where an episode starts, what the agent may do, and what it sees, while the underlying simulator, tasks, and human-built verifier stay untouched. An LLM designer called EnvRigger writes those wrappers automatically against flaws it diagnoses in the policy’s own rollouts. Across five benchmarks in four domains, skills mined this way gain up to 9.0 points on held-out tasks with 9.8% fewer execution steps. Is it deployable? Yes, if you already run an agent eval loop. EnvHarness ships as Apache-2.0 Python with reproduction drivers for six environments. A new benchmark joins by implementing one interface (reset / step / observe / evaluate / get_env_state / save_state / from_state); nothing downstream changes. The hard prerequisite is a resettable environment, which rules out live user accounts and physical robots. https://arxiv.org/pdf/2608.19880 Environments that stop teaching LLM agents now learn less from curated text and more from interactive environments. Those environments are hand-built and static: they behave identically no matter which agent acts or how much it has improved, so they cannot target a policy’s weakness and have nothing left to teach once solved. The usual answer is generating more environments. The EnvHarness paper names two costs: generation pipelines are domain-specific and do not transfer, and LLM-written verifiers must be over-generated and heavily filtered without ever being fully trustworthy. Wrapping, not authoring The research team proposes the opposite move. An agent harness makes a frozen LLM capable through plug-in tools, memory and skills. EnvHarness applies that idea to the other side of the loop, wrapping a frozen environment in plug-in components that operate strictly through the standard reset() / step() interface. Formally, a component is a transformation E’ = w(E) that rewrites the state, action, observation and transition terms. The reward term is deliberately left out. Because no intervention reaches the simulator backend, every reshaped task keeps its original, human-built verifier, and because nothing touches benchmark-specific code, one implementation covers every domain. Three components ship, and they compose freely: Stage replays a fixed action list after reset(), so the episode starts somewhere else. Hiding the target mug in a closed drawer forces search instead of reach. Contract installs per-step hooks on the action, transition and observation axes: block an action, rewrite a response, truncate an observation. Chain composes a second environment into the same episode under a shared step budget, with the composite verdict being the conjunction of both verifiers. EnvRigger: the designer loop Components are policy-agnostic; choosing them is not. EnvRigger treats the policy as a black box and runs four stages: it observes five baseline rollouts, diagnoses a systemic flaw, writes components as real Python, and validates on five fresh rollouts. Unsolvable and trivially solvable candidates are both rejected, with up to five revision rounds per task. Generated hooks compile in an isolated subprocess, so a bad mutation becomes a recorded trace rather than a dead run. Performance Across ALFWorld, WebArena, SWE-bench Verified, OfficeQA and SpreadsheetBench, skills mined with ReasoningBank-style induction beat both controls on untouched held-out tasks. ALFWorld average rises from 62.4 to 68.3 against original-environment skills, with +9.0 points on the out-of-distribution split. SWE-bench Verified resolved rate moves 49.88 → 52.58 while average steps fall 55.01 → 49.61, the paper’s 9.8% efficiency claim. On SpreadsheetBench and WebArena, skills from unmodified environments land below the no-skill baseline; reshaping is what makes mining worthwhile. Against domain-specific generators, EnvHarness beats SWE-smith by 2.46 points with 5.11 fewer steps. Under GRPO on Qwen3-8B-base, RL in reshaped environments beats RL in the originals on three of four metrics (ALFWorld in-distribution 81.4 → 87.9), with a small regression on the OOD split (89.6 → 88.8). Environment scaling reaches 54.79 at 300 environments versus 52.13 for originals and 50.37 for generated ones, because the designer co-evolves each batch against the current policy. And asked to steer per-task success rate into [0.4, 0.6], in-band coverage rises from 6% to 80%. Key Takeaways EnvHarness wraps frozen environments through reset()/step() only, so verifiers stay human-built. Three components — Stage, Contract, Chain — cover start state, interaction rules, and episode composition. EnvRigger diagnoses policy flaws from rollouts and writes targeted wrappers, validating on fresh rollouts. Gains hold across five benchmarks: +9.0 points OOD on ALFWorld, 9.8% fewer steps on SWE-bench Verified. Apache-2.0 code is live; the cost is designer tokens and a hard requirement for resettable environments. Check out the Paper, GitHub Repo and Project Page. Also, feel free to follow us on Twitter and don’t forget to join our 150k+ML SubReddit and Subscribe to our Newsletter. Wait! are you on telegram? now you can join us on telegram as well. Need to partner with us for promoting your GitHub Repo OR Hugging Face Page OR Product Release OR Webinar etc.? Connect with us The post Google AI Introduces EnvHarness: A Programmable Layer That Turns Static Agent Environments Into Adaptive Training Worlds appeared first on MarkTechPost.