iCreatAI News
Back to main site
Skip to content
Only First party#Research
  1. Microsoft Research78

    Lazare, veni foras: Reprogramming Dormant Neurons to Backdoor AI Agents

    Microsoft researchers describe "Lazarus," a backdoor technique that reprograms fewer than 10 dormant neurons in an LLM, without training, so an agent built on it behaves normally and passes standard evals until a trigger—such as a date change—causes it to download and run attacker code with whatever access it has. The attack needs only write access to a checkpoint, so internally developed models are equally exposed, and reliable detection requires diffing against original weights, making provenance and weight integrity critical. Microsoft Foundry ties open-weight models to their publishers and scans surrounding containers, and Microsoft is developing AI-powered weight scanning, a red-teaming harness, and cryptographic model signing.

    Why it matters: It offers a concrete, researcher-first account of a low-cost weight-space backdoor and the provenance and weight-auditing practices it implies for anyone deploying agents.

  2. Artificial Analysis Articles60

    Multimodal Knowledge Agent (MKA) benchmark: testing knowledge work capabilities with complex visual files

    Artificial Analysis partnered with Google DeepMind to launch the Multimodal Knowledge Agent (MKA) benchmark, comprising 200 private real-world visual tasks across 10 industries in which agents work from dense visual files like architectural drawings, engineering schematics, and clinical scans to produce deliverables such as slide decks, spreadsheets, PDFs, Word documents, and marked-up images. Tasks are graded against expert rubrics averaging ~70 weighted criteria split into Regression and Climb checks, with the headline Climb WSR score, run in the open-source Stirrup agent harness with up to 250 turns and graded by GPT-6 Luna (medium) as an agentic judge. Claude Fable 5.2 (max) leads with 57% Climb WSR, followed by Gemini 4 Carbon (high) at 54% and Claude Opus 5.5 (max) at 53%, while Claude Fable 5.2 averages $31.97 per task versus $6.29 for Gemini 4 Carbon and $1.11 for GPT-6.1 Sol.

    Why it matters: The results offer a useful cost-performance and per-vertical breakdown for anyone choosing models for agentic knowledge work over complex visual files.

  3. Baseten Blog69

    How Baseten and Genspark trained a deep research model to outperform Opus 5.5

    Genspark worked with Baseten Applied Research to post-train Qwen3.5-397B into a specialized Deep Research model that scores 0.759 on Genspark's quality grader, up from 0.612 for the base model and above Opus 5.5 (0.738), Sonnet 5.5 (0.742), GPT-5.6 Sol (0.637), Kimi K3 (0.724) and DeepSeek V4 Pro (0.675). The recipe splits the task into research, trained with reinforcement learning against a shaped semi-verifiable reward, and synthesis, trained mainly with supervised fine-tuning on high-quality reports written from the model's own research trajectories, followed by end-to-end RL. The model crawls 5× more sources per query than the base model and runs with NVFP4 weights on 2×B200 GPUs per replica, now serving as the default for Genspark's Deep Research agent in production.

    Why it matters: A detailed, reproducible account of post-training an open model for a judged production workload, useful for teams weighing RL versus SFT and reward shaping.

  1. Google DeepMind83

    Introducing AlphaProof Code: a breakthrough in verifying complex software at depth

    Google DeepMind researchers used AlphaProof Code, a multi-LLM agentic framework combining human expertise with AI agents, to formally verify key security properties of the actual code underlying the Signal protocol, generating over six million lines of Lean. The verification took four weeks, and when two previously fixed bugs were reintroduced into an isolated version of Signal's code, the system produced machine-checked proofs that the buggy versions were insecure. DeepMind is open-sourcing its formal verification artifacts.

    Why it matters: It reports a claimed first in formally verifying a widely deployed cryptographic protocol's implementation, with concrete details on method, timeline and bug-detection tests.

  1. vLLM Blog72

    Zero Mismatch, Simple Async RL: REINFORCE Zero, DeepSeek-V4.1 and Kimi K3.1

    The post argues that asynchronous RL's rollout-to-learner logprob gap splits into smoothly varying weight lag, which unclipped importance sampling can correct, and fluctuating kernel mismatch, which persists despite IS; aligning kernels first lets REINFORCE Zero train stably for 20,000 updates, learn at policy lag 64, and reach 2.40× the training-step throughput of synchronous training with aligned kernels. It reports zero mismatch (bit-for-bit logprob agreement) for DeepSeek-V4.1-Flash and Kimi K3.1 on day 0, and for DeepSeek-V4-Flash and Kimi K3.1 on AMD MI355X. Code and recipes are in Inferact/zero-mismatch-rl, with the paper at hijkzzz/REINFORCE-Zero and training code in NVIDIA-NeMo/labs-molt.

    Why it matters: It offers a concrete diagnosis of why async RL drifts and a minimal recipe, backed by throughput, lag-64 and cross-vendor alignment results.

  2. METR Blog78

    Independent Assessment of AI R&D Acceleration at Anthropic in October 2026

    METR conducted a high-access evaluation at Anthropic from August 12 to October 29, 2026, concluding that Anthropic is likely improving model capabilities on well-specified tasks at around twice its March 2024–October 2025 rate, with AECI benchmark scores growing at an estimated 1.9x the pre-October 2025 rate (95% CI: 1.5–2.8x). METR estimates that a majority (perhaps ~80%) of the current acceleration is due to AI, found no clear bottlenecks to continued progress over the next 12 months, and cautioned that Anthropic redacted most of the evidence it found most informative.

    Why it matters: It offers a rare public accounting of how frontier AI R&D acceleration is measured, including explicit uncertainty ranges and disclosure of what redactions prevented the public from judging.

  3. Google Research71

    Detecting signs of hypertension using wearable sensors

    Google researchers describe AI-PPG-ACC-HTN, a multimodal AI system that uses smartwatch PPG and accelerometry data, without cuff calibration, to detect a persistent hypertensive state by aggregating 15-second waveform embeddings over multiple days alongside height. Validated against 24-hour ambulatory blood pressure monitoring in a 196-participant, five-site US study, it reached 65.8% sensitivity and 90.0% specificity for initial screening and 78.4% sensitivity, 90.9% specificity and 96.7% PPV for confirmatory testing among 48 participants with elevated office BP. The authors frame it as early feasibility research for passive opportunistic screening, not a replacement for clinical cuffs or a standalone diagnosis.

    Why it matters: It offers a concrete, clinically benchmarked example of cuffless hypertension screening and its limits, useful for readers tracking AI in consumer health devices.

  1. Anthropic Research66

    Investigating vault particles with Claude

    Anthropic's biology research lab reports the first 3D structure of a vault particle from a prokaryotic cell—a bacteriophage-derived vault built from 96 copies of a single protein, L05, solved by cryo-EM. Under researchers' direction, Claude searched protein databases, designed DNA constructs, optimized a size-exclusion chromatography isolation method that cut purification from about three days to roughly 90 minutes, and processed microscopy data, yielding a first structural model in 34 hours. The team also engineered altered and enlarged vault variants, and notes the work demonstrates using Claude across every step of fundamental biology research.

    Why it matters: Describes a concrete AI-assisted structural biology result, including the methods, speedups, and limitations the team encountered.

  1. Hugging Face Blog60

    Same Model, Different Harness: CUGA, Hermes, and Deep Agents — When to Pick CUGA?

    IBM ran CUGA, Hermes, and Deep Agents on the same open-weight model (GLM-5.3-flash) across 90 AppWorld tasks, and task completion ranged from 30% to 81% depending only on the harness, with CUGA at 81.1% (73/90) versus Hermes at 46.7% and Deep Agents at 30.0%. CUGA is IBM's Apache 2.0 harness for enterprise API work, adding runtime-enforced policies (intent guards, tool approvals, ToolGuards, output formatting) and deployment inside sovereign or air-gapped environments. The authors note the 90-task sample, non-identical setups, and absence of policy testing as limits, and say CUGA is not the right fit for terminal coding or personal chat assistants.

    Why it matters: Useful for teams choosing an agent harness, since it isolates harness effects with a fixed model and shows where runtime policy enforcement fits enterprise needs.

  1. Replit Blog60

    ViBench 2.0: A Benchmark for Long-Horizon Software Building

    Replit releases ViBench 2.0, a long-horizon software development benchmark in which an agent builds a product from an empty project and then adds up to eighteen features on top of its own code, spanning seventeen products, 241 feature requests and 544 test plans. In Fireworks' official runs, GPT-6.1 Sol leads with a pass@1 of 65%, followed by GPT-6 Astra at 60% and Opus 5.5 at 49%, and no open-weight model passes a full build. Eight of the products are public on GitHub in the Harbor format, and results can be rerun from release v2.0.6 on the OpenHands-based harness.

    Why it matters: It offers concrete pass rates, costs and failure modes for frontier models on multi-feature product building, with a public, rerunnable harness.

  1. Hugging Face Blog62

    Turning Database Grading into RL Rewards: Post-Training Agents with ThinkingBox

    Microsoft and Hugging Face describe using ThinkingBox's executable terminal-database-state checks as binary rewards to fine-tune Qwen3.8-27B with GRPO for 50 steps on an offline-selected pool of 187 tasks, raising pass@1 from 51.70% to 60.78% and tasks passing all 20 attempts from 38 to 107 on the 507-task ThinkingBox-Bench, while pass@20 dipped slightly from 453 to 446. The post details offline policy-relative difficulty selection via Jeffreys posteriors, the reward loop, cost-per-dependable-task falling from $24.36 to $8.00, and releases the training stack in microsoft/thinkingbox-training.

    Why it matters: It offers a concrete, reproducible recipe for converting stateful agent grading into RL rewards, plus an honest account of the repeatability-versus-coverage tradeoff.

  1. Google Research65

    Empowering frontline maternal care with AI-assisted ultrasound

    Google researchers evaluated an AI model that estimates gestational age from standardized "blind sweep" ultrasounds performed by novice operators, reporting a mean absolute error of 4.18 days versus 4.48 days for expert sonographers across cohorts in Chicago and Nairobi. The model, minimally tuned on 120 Chicago participants and tested with no Nairobi data, achieved MAEs of 4.10 and 4.27 days respectively, as published in JAMA Network Open.

    Why it matters: It offers a concrete, peer-reviewed data point on whether AI-assisted ultrasound can generalize across hardware and clinical settings without expert sonographers.

  2. Microsoft Research78

    Solving a 31-year-old grand challenge in mathematics – the KLS conjecture

    Microsoft researcher Xinzhi Zhang and a collaborator proved the Kannan–Lovász–Simonovits (KLS) conjecture, proposed in 1995, showing that for log-concave distributions the best hyperplane cut is within a universal constant of the best cut of any shape regardless of dimension. The proof emerged from a human-directed agentic process exploring more than 100 candidate approaches, and was followed by independent parallel proofs and a Lean 4 formalization. The authors say unguided attempts with leading AI models using a similar token budget fell short, while the success depended on both AI capabilities and researchers' mathematical judgment.

    Why it matters: It documents a major open problem's resolution and offers a concrete account of how human-guided AI search contributed to a mathematical proof.

  1. Google Research71

    Grounding AI agents in biological reality: Building blocks for autonomous wet labs

    Google researchers describe three projects aimed at connecting AI agents to real wet-lab experimentation: the VERID framework, a semantic digital twin of a lab workcell that tracks 3D state, runs fast virtual dry runs, blocks unsafe commands and logs synchronized video and actions; closed-loop aptamer design over three design-test cycles that raised on-target hit rates to 33.5% for myoglobin and 42.1% for α-thrombin; and agent-guided epitope excision for de novo binder design against FAP, which accelerated each folding cycle more than eightfold and, in a human–agent comparison at equal compute, nearly doubled the in-silico hit rate of a scientist working alone.

    Why it matters: It offers a concrete, example-rich account of how agents are being tied to physical lab hardware and closed experimental loops, including failure modes and safety safeguards.

  1. Google Research65

    Putting Earth AI to work: Geospatial agents for our planet's biggest challenges

    Google shares a technical report on its experimental Geospatial Reasoning agent, which connects Gemini to Earth AI foundation models and planet-scale data, and reports evaluations across four benchmarks, including two newly released ones (GeoKTools and GeoExpertQueries). The post describes early partner testing with organizations including WHO AFRO, People in Need, GiveDirectly, the Milken Institute and WPP, and announces expanded early access via Ask Google Earth and an experimental agent-to-agent API.

    Why it matters: It offers a concrete look at how a planning-based geospatial agent is evaluated and applied to real public health, disaster and infrastructure workflows.

  1. Google Research71

    Does better work always mean better workers?

    A three-month field experiment with 133 patent lawyers found that access to a Google Labs AI patent-writing assistant raised drafting scores by 0.34 SD at 10 days and 0.38 SD at 90 days, with gains concentrated in fewer poor scores rather than more excellent ones. On a later redlining task where AI was not allowed, the benefit appeared only for senior lawyers (0.45 SD), while juniors showed no discernible improvement and their scores bifurcated toward more very low and more good results. The authors argue that improving output and building durable professional judgment are different goals that may be in tension for junior professionals.

    Why it matters: Offers a rare blinded, months-long field test that separates AI-assisted output gains from unassisted skill development, with direct implications for how firms train junior professionals.

  1. GitHub Blog · AI & ML61

    ReviewBench: An open benchmark for AI code review

    GitHub introduces ReviewBench, an open benchmark for AI code review built from analysis of 103.9 million GitHub pull requests, containing 219 public pull requests across 19 languages with a multi-source golden set labeled for severity and category. It reports grounded and augmented precision, recall, and F1 metrics, uses Claude Sonnet 5 as the LLM grader under a published rubric, and had its ground-truth findings independently re-labeled by senior engineers with 96.6% agreement. GitHub says ReviewBench's offline predictions for a Copilot code review multi-model ensemble matched a later A/B test, which showed an 8.0% rise in addressed rate, 13.6% in recall, 61% in comment volume, and an 8.0% drop in cost per review.

    Why it matters: It offers a concrete, reproducible methodology and public dataset for comparing AI code reviewers, plus evidence on how offline scores tracked a production experiment.

  1. Berkeley AI Research60

    From CUDA to MLX: How K-Search Brings Decades of Kernel Expertise to Apple Silicon

    IBM Research extended K-Search, an evolutionary LLM-driven kernel search framework from UC Berkeley Sky Lab, with a native MLX backend and a structured CUDA-to-MLX translation layer that maps CUDA primitives, constraints and optimization patterns onto Apple Silicon equivalents. On Apple Silicon the approach reached 0.97x the speed of the native MLX Attention kernel and up to a 20x prefill speedup over the community mlx-lm implementation on the Mamba SSM kernel, attributed largely to adding a parallel scan. The authors say the bottleneck was not the LLM's ability to write Metal code but the quality of context and constraints supplied to it.

    Why it matters: Describes a concrete method for transferring CUDA kernel expertise to non-NVIDIA hardware, with reported benchmark numbers and an open-source reproduction path.

You have reached the end