The arXiv preprint "CyberBinGym: Can AI Agents Turn Binary Patches into Real Exploits?" (arXiv:2611.10723, submitted 7 Nov 2026) introduces a benchmark of 100 tasks derived from real vulnerabilities in 53 widely used projects. Agents receive stripped vulnerable and patched binaries "without source code or vulnerability information". According to the abstract, "The strongest evaluated agent recovers the target vulnerability in 88% of tasks and exploits it in 20%." Providing the patched binary "raises its exploit rate from 10% to 25% on a matched 20-task subset". The authors write that these results "demonstrate that current agents can turn released binary patches into working exploits in a controlled setting, although exploitation remains unreliable." These are the authors' own benchmark results in a controlled setting, not independent evaluations.
On 10 Nov 2026 Microsoft Research published "Lazare, veni foras: Reprogramming Dormant Neurons to Backdoor AI Agents", describing a backdoor technique it calls "Lazarus" that "can reprogram dormant neurons in an AI model, causing an otherwise normal AI agent to execute malicious commands when a hidden trigger appears." The Microsoft Security blog post of the same day says the method edits "fewer than 10 neurons to inject a short payload" and that the sleeper agent "passes all evaluations" until a trigger string appears. The Microsoft Research post says the backdoor needs no training run and is "difficult to detect through normal testing"; the Security blog says the attack "needs inference-only compute." Microsoft says it "has not observed any threat actors weaponizing this technique." The Microsoft Security blog post also says "Claude Code with Opus 4.5 discovered most of this threat autonomously." These are company-reported research claims; the paper's experimental figures were not checked by Scout.
Meta's developer blog introduces Muse Ember in Muse Code and on the Meta Model API. The post says: "Built for fast, agentic execution at scale, Muse Ember handles coding tasks, tool calls, and document analysis," and that Muse Ember is "optimized for low price-per-task at $0.10 input / $0.50 output per million tokens for up to 200k input tokens". For prompts above 200k input tokens the post gives $0.30 input / $1.50 output. The post's comparison table, which Meta presents, names Claude Haiku 5.5 (max) and GPT-6 Luna (max) and includes an "Artificial Analysis Index" row and a "Personal Superintelligence Index", which the post describes as an "Internal composite index of personal and knowledge work". The post shows no publication date. A secondary report says Meta released it on Monday, 9 November 2026. The benchmark results and the partner statements in the post are company-presented and not independently verified.
On 8 Nov 2026 Malbear Labs published a report on "Operation Snowball", an unsecured open directory that held an autonomous AI intrusion harness. The report describes "more than two weeks of largely unsupervised intrusion activity" against a European telecommunications provider and organisations hosted on its infrastructure. The harness was "built around DeepSeek Flash", and the report says it was "directed by a Chinese-speaking operator". Across more than 70 recovered sessions, the report lists "Hundreds of assets enumerated, most confirmed live" and "Millions of patient records accessed via a TrakCare Oracle instance with default database credentials". The post presents these as what the recovered records show. It describes no independent verification, and these figures are not verified by the affected organisations. The telecom is not named. The 13 organisations are targets the agent compiled from public RIPE registry records, not confirmed breaches. The report does not name a threat group or confirm the operator's identity.
On 10 Nov 2026 Google announced Gemini 4 Carbon, available through the Gemini API and Google AI Studio, Gemini for business, and to Google AI Ultra subscribers in Google Antigravity, the Gemini app and AI Mode. Google calls Carbon its strongest model, with frontier performance in cybersecurity capabilities, and says Gemini 4 Argon remains the ideal choice for the majority of software engineering and knowledge work. The introductory price is $2 per million input and $10 per million output tokens; from 30 Nov 2026 it rises to $4 and $20. Google reports that Carbon ranks first on CursorBench 4.0, Perplexity's WANDR and Zapier's AutomationBench. In its own safety tests, Carbon blocked 99% of attack trajectories from an adaptive red-teaming agent, and the Gray Swan automated cyberattack success rate fell to 2.17% from 9.86% for Argon. Google also reports 100% recall and 95% precision on its internal benchmark, Project Fortify, and that Wiz found 0-day vulnerabilities in under two hours. All figures are company-reported or partner-reported and have not been independently checked.
On 10 Nov 2026 Anthropic announced that it had launched the roadshow for a proposed initial public offering of its Class A common stock. The registration statement has been filed with the SEC but is not yet effective. The offering is 825,000,000 shares: Anthropic offers 801,351,354 and certain existing stockholders offer 23,648,646, with a 30-day option for underwriters to buy up to 123,750,000 more. Anthropic says it receives no proceeds from the shares sold by stockholders. The expected price range is $70.00–$78.00 per share. Anthropic has applied to list on the Nasdaq Global Select Market under the symbol ANT, with Goldman Sachs & Co. LLC and Morgan Stanley & Co. LLC named as the sources for the preliminary prospectus. The announcement does not state a valuation, pricing date or listing date. Scout's own arithmetic, not in the post: at the stated range, the 801 million company shares would be about $56–62.5 billion gross, before fees. All terms are company-stated and may change before pricing.
32 published signals in the current snapshot. 32 signals: 31 reviewed by editors, 1 published automatically after an AI check. 4 sources are monitored for new items; each is checked before it is published. This is a curated selection, not a complete record of global events.
A global perspective.
Published signals grouped by editorial region. These are navigation categories, not a claim of complete regional monitoring.
Regions not listed have no published signals in this snapshot; that is a coverage limit, not evidence of inactivity. 24 published signals have no established country and are listed without a region.
