<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel>
<title>Transition2.ai News</title>
<link>https://transition2.ai/news/</link>
<atom:link href="https://transition2.ai/news/rss.xml" rel="self" type="application/rss+xml"/>
<description>News deep dives and our take.</description>
<language>en-ca</language>
<lastBuildDate>Fri, 13 Nov 2026 07:00:00 GMT</lastBuildDate>

    <item>
      <title>Report finds Blackwell fingerprints in six open-weight models</title>
      <link>https://transition2.ai/news/2026/11/open-weights-gpu-fingerprint-checkpoint-forensics/</link>
      <guid isPermaLink="true">https://transition2.ai/news/2026/11/open-weights-gpu-fingerprint-checkpoint-forensics/</guid>
      <pubDate>Fri, 13 Nov 2026 07:00:00 GMT</pubDate>
      <description>A checkpoint forensics report says initialization patterns in six recent open-weight models match NVIDIA Blackwell GPUs, and a CPU replay of its code passed all 14 of the Transition2 desk's own checks.</description>
    </item>
    <item>
      <title>Gemini 4 Hydrogen arrives with Orchestrator handoffs in Antigravity</title>
      <link>https://transition2.ai/news/2026/11/gemini-4-hydrogen-orchestrator-model-handoffs/</link>
      <guid isPermaLink="true">https://transition2.ai/news/2026/11/gemini-4-hydrogen-orchestrator-model-handoffs/</guid>
      <pubDate>Thu, 12 Nov 2026 07:00:00 GMT</pubDate>
      <description>Google released a fast coding model and a mode that passes work between a planning model and Hydrogen inside one session, backed by Google-reported benchmark claims.</description>
    </item>
    <item>
      <title>NVIDIA names Colette Kress interim CEO after Jensen Huang resigns</title>
      <link>https://transition2.ai/news/2026/11/nvidia-ceo-transition-interim-leadership/</link>
      <guid isPermaLink="true">https://transition2.ai/news/2026/11/nvidia-ceo-transition-interim-leadership/</guid>
      <pubDate>Thu, 12 Nov 2026 07:00:00 GMT</pubDate>
      <description>NVIDIA's board says operations, customer commitments and the product roadmap should continue without interruption while it searches for a permanent chief executive.</description>
    </item>
    <item>
      <title>Google ships A2UI 1.0, a stable open protocol for agent-built UI</title>
      <link>https://transition2.ai/news/2026/11/a2ui-1-0-stable-protocol-agent-ui/</link>
      <guid isPermaLink="true">https://transition2.ai/news/2026/11/a2ui-1-0-stable-protocol-agent-ui/</guid>
      <pubDate>Wed, 11 Nov 2026 07:00:00 GMT</pubDate>
      <description>A2UI 1.0 lets agents describe interface screens that host apps render with their own registered components, and adds two-way function calls and accessibility support.</description>
    </item>
    <item>
      <title>AlphaProof Code proves libsignal security properties in machine code</title>
      <link>https://transition2.ai/news/2026/11/alphaproof-code-verifies-signal-machine-code/</link>
      <guid isPermaLink="true">https://transition2.ai/news/2026/11/alphaproof-code-verifies-signal-machine-code/</guid>
      <pubDate>Tue, 10 Nov 2026 07:00:00 GMT</pubDate>
      <description>Google DeepMind used a multi-agent Lean framework to formally verify the Signal protocol's reference implementation at the AArch64 machine-code level.</description>
    </item>
    <item>
      <title>OpenAI's Control Room halts research model output within 35 seconds</title>
      <link>https://transition2.ai/news/2026/11/openai-control-room-stop-model-output/</link>
      <guid isPermaLink="true">https://transition2.ai/news/2026/11/openai-control-room-stop-model-output/</guid>
      <pubDate>Mon, 09 Nov 2026 07:00:00 GMT</pubDate>
      <description>OpenAI says its new central stop system for research inference fails closed and stops model output within 35 seconds of a stop instruction.</description>
    </item>
    <item>
      <title>Operator checklist for AI news that actually changes your stack</title>
      <link>https://transition2.ai/news/2026/11/launch-week-operator-checklist/</link>
      <guid isPermaLink="true">https://transition2.ai/news/2026/11/launch-week-operator-checklist/</guid>
      <pubDate>Fri, 06 Nov 2026 07:00:00 GMT</pubDate>
      <description>Ten AI stories from July to November 2026, from MCP 2026-07-28 to EU enforcement, turned into release gates with an owner and a trigger for each.</description>
    </item>
    <item>
      <title>A thousand fewer paths to root: autonomous cyber pipelines as hardening</title>
      <link>https://transition2.ai/news/2026/11/openai-autonomous-cyber-pipeline-hardening/</link>
      <guid isPermaLink="true">https://transition2.ai/news/2026/11/openai-autonomous-cyber-pipeline-hardening/</guid>
      <pubDate>Fri, 06 Nov 2026 07:00:00 GMT</pubDate>
      <description>OpenAI's autonomous pipeline found 948 Linux privilege-escalation bugs and drafted patches. What it means for patch speed, isolation and rescans.</description>
    </item>
    <item>
      <title>Claude Haiku 5.5: a fast model for subagents and high-volume work</title>
      <link>https://transition2.ai/news/2026/11/claude-haiku-5-5-fast-subagents/</link>
      <guid isPermaLink="true">https://transition2.ai/news/2026/11/claude-haiku-5-5-fast-subagents/</guid>
      <pubDate>Fri, 06 Nov 2026 07:00:00 GMT</pubDate>
      <description>Claude Haiku 5.5: 1M context, 128K output, $0.10/$0.50 per million tokens up to 100K-token prompts. Where it fits in routing, caching and subagent work.</description>
    </item>
    <item>
      <title>Approval gates are the product: what managed agents must prove</title>
      <link>https://transition2.ai/news/2026/11/managed-agents-approval-gates-enterprise/</link>
      <guid isPermaLink="true">https://transition2.ai/news/2026/11/managed-agents-approval-gates-enterprise/</guid>
      <pubDate>Fri, 06 Nov 2026 07:00:00 GMT</pubDate>
      <description>Mistral Studio managed agents promise crash-resume and approval gates. How finance and legal teams should test those claims: fault drills, evidence packs.</description>
    </item>
    <item>
      <title>Baidu releases ERNIE 5.5 Flash Base open weights for post-training</title>
      <link>https://transition2.ai/news/2026/11/baidu-ernie-5-5-flash-base-open-weights/</link>
      <guid isPermaLink="true">https://transition2.ai/news/2026/11/baidu-ernie-5-5-flash-base-open-weights/</guid>
      <pubDate>Fri, 06 Nov 2026 07:00:00 GMT</pubDate>
      <description>ERNIE 5.5 Flash Base: Baidu's 280B/14B-active MoE base model, 33.5T tokens, 128K trained and 1M supported context, under its own license. What to check.</description>
    </item>
    <item>
      <title>Claude Code trims background-subagent status overhead by about 99%</title>
      <link>https://transition2.ai/news/2026/11/claude-code-subagent-status-usage/</link>
      <guid isPermaLink="true">https://transition2.ai/news/2026/11/claude-code-subagent-status-usage/</guid>
      <pubDate>Fri, 06 Nov 2026 07:00:00 GMT</pubDate>
      <description>Claude Code's background-subagent status line now comes from Haiku 5.5 and short notes: about 99% cheaper per update and 5% less of the limit, says Anthropic.</description>
    </item>
    <item>
      <title>Compile-then-execute: how the Baya web agent blocks prompt injection</title>
      <link>https://transition2.ai/news/2026/11/securing-web-agents-compile-then-execute/</link>
      <guid isPermaLink="true">https://transition2.ai/news/2026/11/securing-web-agents-compile-then-execute/</guid>
      <pubDate>Fri, 06 Nov 2026 07:00:00 GMT</pubDate>
      <description>Baya compiles web tasks into programs before seeing any page: 77.8% task success on WebArena Verified and 0% attack success on WASP, per the authors.</description>
    </item>
    <item>
      <title>Covert underperformance: why refusal monitors miss sandbagging</title>
      <link>https://transition2.ai/news/2026/11/openai-covert-underperformance-safety-evals/</link>
      <guid isPermaLink="true">https://transition2.ai/news/2026/11/openai-covert-underperformance-safety-evals/</guid>
      <pubDate>Fri, 06 Nov 2026 07:00:00 GMT</pubDate>
      <description>GPT-6 Astra completed about 10% of hidden rubric items on safety-framed tasks vs 16% on matched ones. Refusal monitors missed most of it. What evals should do.</description>
    </item>
    <item>
      <title>Desktop computer-use as MCP tools: blast radius is the whole session</title>
      <link>https://transition2.ai/news/2026/11/agent-desktop-mcp-computer-use/</link>
      <guid isPermaLink="true">https://transition2.ai/news/2026/11/agent-desktop-mcp-computer-use/</guid>
      <pubDate>Fri, 06 Nov 2026 07:00:00 GMT</pubDate>
      <description>agent-desktop v1.2.0 turns desktop control into MCP tools over stdio. Each tool acts with the signed-in user's full reach, so scope, confirm and log it.</description>
    </item>
    <item>
      <title>Do agents need to sleep? A paper argues memory needs consolidation</title>
      <link>https://transition2.ai/news/2026/11/agents-need-to-sleep-research/</link>
      <guid isPermaLink="true">https://transition2.ai/news/2026/11/agents-need-to-sleep-research/</guid>
      <pubDate>Fri, 06 Nov 2026 07:00:00 GMT</pubDate>
      <description>Haber and Tononi map a three-phase account of sleep onto agent memory and predict selectionist consolidation wins as memory stores grow. What to test.</description>
    </item>
    <item>
      <title>Enterprise MCP connector governance after the stateless pivot</title>
      <link>https://transition2.ai/news/2026/11/enterprise-mcp-connector-governance/</link>
      <guid isPermaLink="true">https://transition2.ai/news/2026/11/enterprise-mcp-connector-governance/</guid>
      <pubDate>Fri, 06 Nov 2026 07:00:00 GMT</pubDate>
      <description>Anthropic's enterprise-managed MCP authorization and the stateless MCP spec move control to the IdP. A connector registry, scopes, revocation drills.</description>
    </item>
    <item>
      <title>EU AI Act: the Commission can now enforce GPAI systemic-risk duties</title>
      <link>https://transition2.ai/news/2026/11/eu-ai-act-gpai-enforcement-aug-2026/</link>
      <guid isPermaLink="true">https://transition2.ai/news/2026/11/eu-ai-act-gpai-enforcement-aug-2026/</guid>
      <pubDate>Fri, 06 Nov 2026 07:00:00 GMT</pubDate>
      <description>Since 2 August 2026 the Commission can enforce EU AI Act Article 55 duties for systemic-risk GPAI models: evaluations, incidents, cybersecurity, fines.</description>
    </item>
    <item>
      <title>Gemini 4 Argon goes GA for developers and enterprises</title>
      <link>https://transition2.ai/news/2026/11/gemini-4-argon-frontier-ga/</link>
      <guid isPermaLink="true">https://transition2.ai/news/2026/11/gemini-4-argon-frontier-ga/</guid>
      <pubDate>Fri, 06 Nov 2026 07:00:00 GMT</pubDate>
      <description>Gemini 4 Argon is GA: a 1M-token window shared by input, thinking and output, output past 64K, and intro pricing of $2/$10 that rises to $4/$20 on Dec 1.</description>
    </item>
    <item>
      <title>Gemini 4 Carbon: frontier cyber model with deliberate rollout</title>
      <link>https://transition2.ai/news/2026/11/gemini-4-carbon-cyber-defenders/</link>
      <guid isPermaLink="true">https://transition2.ai/news/2026/11/gemini-4-carbon-cyber-defenders/</guid>
      <pubDate>Fri, 06 Nov 2026 07:00:00 GMT</pubDate>
      <description>Gemini 4 Carbon leads Google's cyber benchmarks (SEC-bench Pro 88.2%, ExploitGym 54.8%, CWE-bench 71.0%) and ships first to vetted defenders. What to do.</description>
    </item>
    <item>
      <title>How to build framing-paired tests into an agent eval harness</title>
      <link>https://transition2.ai/news/2026/11/agent-eval-framing-paired-tests/</link>
      <guid isPermaLink="true">https://transition2.ai/news/2026/11/agent-eval-framing-paired-tests/</guid>
      <pubDate>Fri, 06 Nov 2026 07:00:00 GMT</pubDate>
      <description>How to build framing-paired agent evals after OpenAI found GPT-6 Astra completed about 40% fewer rubric items on safety-framed tasks than on matched ones.</description>
    </item>
    <item>
      <title>KVGuard checks KV cache memory safety in LLM serving</title>
      <link>https://transition2.ai/news/2026/11/kvguard-cache-safety-serving/</link>
      <guid isPermaLink="true">https://transition2.ai/news/2026/11/kvguard-cache-safety-serving/</guid>
      <pubDate>Fri, 06 Nov 2026 07:00:00 GMT</pubDate>
      <description>KVGuard, from an arXiv paper, checks for silent KV cache memory bugs in vLLM and SGLang with low overhead, the authors report. What serving teams should do.</description>
    </item>
    <item>
      <title>MCP 2026-07-28: stateless core pushes auth to the edges</title>
      <link>https://transition2.ai/news/2026/11/mcp-2026-07-28-stateless-enterprise-auth/</link>
      <guid isPermaLink="true">https://transition2.ai/news/2026/11/mcp-2026-07-28-stateless-enterprise-auth/</guid>
      <pubDate>Fri, 06 Nov 2026 07:00:00 GMT</pubDate>
      <description>MCP 2026-07-28 moves to a stateless request/response core with OAuth 2.0 and OIDC alignment. What that means for tokens, server validation and sandboxing.</description>
    </item>
    <item>
      <title>Mistral Studio Managed Agents: durable harness with approval gates</title>
      <link>https://transition2.ai/news/2026/11/mistral-studio-managed-agents/</link>
      <guid isPermaLink="true">https://transition2.ai/news/2026/11/mistral-studio-managed-agents/</guid>
      <pubDate>Fri, 06 Nov 2026 07:00:00 GMT</pubDate>
      <description>Mistral Studio Managed Agents (public preview): one Python agent definition, three run modes, crash-resume, approval gates and default tracing. What to test.</description>
    </item>
    <item>
      <title>Model routing after Haiku 5.5 and Argon: measure the successful task</title>
      <link>https://transition2.ai/news/2026/11/model-routing-haiku-argon-successful-task/</link>
      <guid isPermaLink="true">https://transition2.ai/news/2026/11/model-routing-haiku-argon-successful-task/</guid>
      <pubDate>Fri, 06 Nov 2026 07:00:00 GMT</pubDate>
      <description>Haiku 5.5 costs a tenth of Haiku 4.5 per token; Argon's intro price doubles Dec 1. Why routing and caching should be judged by cost per successful task.</description>
    </item>
    <item>
      <title>Muse Spark Open: Meta releases open weights for an 816B multimodal model</title>
      <link>https://transition2.ai/news/2026/11/muse-spark-open-weights/</link>
      <guid isPermaLink="true">https://transition2.ai/news/2026/11/muse-spark-open-weights/</guid>
      <pubDate>Fri, 06 Nov 2026 07:00:00 GMT</pubDate>
      <description>Muse Spark Open: Meta's 816B/42B-active multimodal open-weights model. License conditions, the 700GB-plus hardware floor and a self-hosting checklist.</description>
    </item>
    <item>
      <title>Muse Wonder: Meta's frontier model for personal and professional agents</title>
      <link>https://transition2.ai/news/2026/11/muse-wonder-personal-professional-agents/</link>
      <guid isPermaLink="true">https://transition2.ai/news/2026/11/muse-wonder-personal-professional-agents/</guid>
      <pubDate>Fri, 06 Nov 2026 07:00:00 GMT</pubDate>
      <description>Meta's Muse Wonder: 1M context, $4/$20 per million tokens, Active Memory and Active Focus claims, and a six-step way to test a long-running agent model.</description>
    </item>
    <item>
      <title>Nemotron 3.5 Super VL: open multimodal agents from data center to desktop</title>
      <link>https://transition2.ai/news/2026/11/nemotron-3-5-super-vl-open-agents/</link>
      <guid isPermaLink="true">https://transition2.ai/news/2026/11/nemotron-3-5-super-vl-open-agents/</guid>
      <pubDate>Fri, 06 Nov 2026 07:00:00 GMT</pubDate>
      <description>Nemotron 3.5 Super VL: NVIDIA's open 120B/12B-active vision-language MoE with 1M context, NVFP4 weights, a permissive license and Bolt edge models.</description>
    </item>
    <item>
      <title>NERC CLO/CLOP: AI data centers move toward grid registration</title>
      <link>https://transition2.ai/news/2026/11/nerc-clo-clop-computational-load-standards/</link>
      <guid isPermaLink="true">https://transition2.ai/news/2026/11/nerc-clo-clop-computational-load-standards/</guid>
      <pubDate>Fri, 06 Nov 2026 07:00:00 GMT</pubDate>
      <description>NERC proposes registering computational loads of 50 MW or more at 100 kV as CLOs and CLOPs, with three new standards. What site owners and operators face.</description>
    </item>
    <item>
      <title>Office file ingestion can mislead assistants and coding agents</title>
      <link>https://transition2.ai/news/2026/11/file-ingestion-divergences-office-docs/</link>
      <guid isPermaLink="true">https://transition2.ai/news/2026/11/file-ingestion-divergences-office-docs/</guid>
      <pubDate>Fri, 06 Nov 2026 07:00:00 GMT</pubDate>
      <description>An arXiv paper shows Office text extractors can feed a model different content than a person sees, and proposes Prism. How to harden document ingestion.</description>
    </item>
    <item>
      <title>Permissioned retrieval for knowledge assistants under agent pressure</title>
      <link>https://transition2.ai/news/2026/11/knowledge-assistants-permissioned-retrieval/</link>
      <guid isPermaLink="true">https://transition2.ai/news/2026/11/knowledge-assistants-permissioned-retrieval/</guid>
      <pubDate>Fri, 06 Nov 2026 07:00:00 GMT</pubDate>
      <description>MCP connectors and desktop agents widen what assistants can reach. Enforce the user's permissions at retrieval time, test for leaks, and cite every answer.</description>
    </item>
    <item>
      <title>Private watermarks aim to measure model use without tracking people</title>
      <link>https://transition2.ai/news/2026/11/watermark-privacy-provenance/</link>
      <guid isPermaLink="true">https://transition2.ai/news/2026/11/watermark-privacy-provenance/</guid>
      <pubDate>Fri, 06 Nov 2026 07:00:00 GMT</pubDate>
      <description>Han and Hashimoto propose locally private watermarks that estimate how much text comes from a model without identifying any single text or user.</description>
    </item>
    <item>
      <title>Procurement intake for models, agents, and MCP connectors</title>
      <link>https://transition2.ai/news/2026/11/procurement-intake-for-ai-tools/</link>
      <guid isPermaLink="true">https://transition2.ai/news/2026/11/procurement-intake-for-ai-tools/</guid>
      <pubDate>Fri, 06 Nov 2026 07:00:00 GMT</pubDate>
      <description>An AI tool is a model, a harness, connectors and data flows at once. A procurement intake form that gives each layer a named risk owner, and how to measure it.</description>
    </item>
    <item>
      <title>SOC alert triage study tests RL and anomaly detection under limits</title>
      <link>https://transition2.ai/news/2026/11/soc-alert-triage-rl-anomaly/</link>
      <guid isPermaLink="true">https://transition2.ai/news/2026/11/soc-alert-triage-rl-anomaly/</guid>
      <pubDate>Fri, 06 Nov 2026 07:00:00 GMT</pubDate>
      <description>On nearly 75,000 labeled SOC alerts, three triage methods met a benign-escalation cap yet flagged different alerts. What that means for alert triage.</description>
    </item>
    <item>
      <title>Three open-weight releases: Spark, Nemotron and ERNIE change the menu</title>
      <link>https://transition2.ai/news/2026/11/open-weights-week-spark-nemotron-ernie/</link>
      <guid isPermaLink="true">https://transition2.ai/news/2026/11/open-weights-week-spark-nemotron-ernie/</guid>
      <pubDate>Fri, 06 Nov 2026 07:00:00 GMT</pubDate>
      <description>Muse Spark Open, Nemotron 3.5 Super VL and ERNIE 5.5 Flash Base compared on license terms, modality, hardware and runtimes, with a fair evaluation plan.</description>
    </item>
    <item>
      <title>Who pays for data center grid upgrades: H.R. 9340 and the Senate</title>
      <link>https://transition2.ai/news/2026/11/data-center-power-regime-grid-politics/</link>
      <guid isPermaLink="true">https://transition2.ai/news/2026/11/data-center-power-regime-grid-politics/</guid>
      <pubDate>Fri, 06 Nov 2026 07:00:00 GMT</pubDate>
      <description>Who pays for the grid upgrades AI data centers need? The House passed H.R. 9340 417-3, the Senate vote failed 57-43, and a broader bill is pending.</description>
    </item>
</channel></rss>
