iCreatAI News
Back to main site
Skip to content
#Microsoft
  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.

  1. The Decoder89

    Anthropic kicks off IPO roadshow, offering shares at $70 to $78 and targeting a trillion-dollar valuation

    Anthropic is pitching investors on an IPO of 825 million Class A shares priced at $70 to $78, targeting a valuation of up to $1.02 trillion, with the offering plus a $26 billion private placement potentially raising up to $90.4 billion. The company reported Q3 revenue of $16.96 billion, up 47 percent quarter over quarter, alongside a commitment to nearly $900 billion in compute capacity with partners including Amazon and Google. The stock will trade on Nasdaq under the ticker "ANT," with Goldman Sachs and Morgan Stanley leading the underwriting syndicate.

    Why it matters: The piece pairs detailed IPO terms and financials with context on market skepticism and AI infrastructure costs, useful for readers tracking the AI investment cycle.

  1. Hacker News AI78

    Google value plunges by $500B as AI fears rattle Wall Street

    Alphabet lost just over $500bn in market value on Thursday, with its stock hitting a six-month low, after its quarterly update showed AI summaries on Search were affecting ad revenue and profits fell 25% even as total revenue grew 25% to $127.6bn. Microsoft and Amazon also reported this week, with Amazon shares falling 6% after hours on negative free cash flow of $42bn, while Microsoft's shares fell less than 1%. Analysts quoted said the results highlight the challenge of funding massive AI investments while sustaining established businesses, and where AI value will ultimately be captured.

    Why it matters: It captures a major market reaction to AI spending across three tech giants and the analyst debate over whether AI investments are paying off.

  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. 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. Microsoft Research71

    Agent Lightning v1.0: A 3,500-Line Lightweight Agentic RL Framework for Training Agents with Real Harnesses

    Microsoft Research Asia introduces Harnessed Agentic RL, a paradigm in which the same agent harness used in deployment participates directly in reinforcement learning, and open-sources Agent Lightning v1.0, a roughly 3,500-line framework with an API Gateway, Rollout Controller, and Customized Trainer built on verl. Agents run as standard Kubernetes jobs without commercial sandbox services, and Collocated Async RL shares GPUs between rollout and updates, achieving about a 2x end-to-end speedup over synchronous RL. An end-to-end coding agent pipeline raised Qwen3.5-9B from 41.8% to 56.4% Pass@1 on SWE-bench Verified using about 6,000 training samples.

    Why it matters: It offers a concrete, reproducible recipe for training deployed agents in place, with code size, infrastructure, and benchmark numbers stated explicitly.

  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.

You have reached the end