// AI Briefing · November 1, 2026
AI Briefing
Blockchain-analytics firm Chainalysis disclosed that an external attacker using "offensive agentic techniques" gained unauthorized access to its systems and internal tools in October. Cisco Talos found that the VOLTA infostealer sends a profile of each victim's stolen data to an LLM through OpenRouter and asks it for a "target assessment" ranking the victim's assets and security habits. S&P revised Meta's credit outlook from stable to negative, citing "negative cash flow and uncertainty around capital expenditure" plans, though it kept the AA- rating. Users report that Claude Code and Claude Opus 5.5 have started refusing to recreate or clone commercial video games, and some say Claude also blocks harmless reverse-engineering such as reading Windows crash dumps. A federal judge denied xAI's motion to dismiss an NAACP lawsuit alleging that the company ran at least 35 combustion turbines at its Colossus data center near Memphis without air permits. A Washington Post investigation found that Microsoft Copilot cited the Ballot Brief, a largely AI-generated voter-guide site, in 75% of the Post's voter-guide tests, compared with 22% for ChatGPT.
In this video
- 01Chainalysis Says Attackers Using 'Offensive Agentic Techniques' Got Into Its SystemsIndustry
- 02Cisco Talos: VOLTA Infostealer Asks Claude, Kimi or Gemini to Rank Victims by What They're WorthResearch
- 03S&P Gives Meta a Negative Outlook Over AI Spending and Negative Cash FlowIndustry
- 04Developers Say Claude Now Refuses to Clone or Reverse-Engineer Video Games, and Some Routine Work TooProduct
- 05Judge Lets NAACP's Clean Air Act Lawsuit Over xAI's Data Center Turbines ProceedPolicy
- 06Washington Post: Microsoft Copilot Leaned on an Error-Prone, AI-Built Voter Guide Days Before the MidtermsPolicy
The stories
Chainalysis Says Attackers Using 'Offensive Agentic Techniques' Got Into Its Systems
Blockchain-analytics firm Chainalysis disclosed that an external attacker using "offensive agentic techniques" gained unauthorized access to its systems and internal tools in October. The accessed data includes user IDs, names, email addresses and organizational role labels for users on its platform. Chainalysis says it has found no evidence that investigation content, credentials or session tokens were accessed. It is one of the first breaches a company has publicly attributed to AI-agent-driven hacking, at a firm whose customers include law enforcement agencies.
Cisco Talos: VOLTA Infostealer Asks Claude, Kimi or Gemini to Rank Victims by What They're Worth
Cisco Talos found that the VOLTA infostealer sends a profile of each victim's stolen data to an LLM through OpenRouter and asks it for a "target assessment" ranking the victim's assets and security habits. Claude Opus 4.6 is the default model. If a model refuses, the malware falls back to Kimi K2 and then Gemini 2.5 Flash. Talos says this shows criminals expect safety refusals and design around them by switching to more permissive models, turning commercial AI into a triage tool for stolen data.
S&P Gives Meta a Negative Outlook Over AI Spending and Negative Cash Flow
S&P revised Meta's credit outlook from stable to negative, citing "negative cash flow and uncertainty around capital expenditure" plans, though it kept the AA- rating. S&P projects Meta's capex at $140B in 2026, $225B in 2027 and $252B in 2028, with negative free operating cash flow of $22B, $71B and $39B over the same years. S&P also said Meta has fewer ways to make money from its AI investments than peers that can rent out compute. The warning brings bond markets into the debate over whether Big Tech's AI buildout can pay for itself.
Developers Say Claude Now Refuses to Clone or Reverse-Engineer Video Games, and Some Routine Work Too
Users report that Claude Code and Claude Opus 5.5 have started refusing to recreate or clone commercial video games, and some say Claude also blocks harmless reverse-engineering such as reading Windows crash dumps. Anthropic has not confirmed or explained the change. The reports point to AI labs tightening their models to limit legal exposure from game and software makers, and the collateral refusals are pushing some developers toward open-weight models.
Judge Lets NAACP's Clean Air Act Lawsuit Over xAI's Data Center Turbines Proceed
A federal judge denied xAI's motion to dismiss an NAACP lawsuit alleging that the company ran at least 35 combustion turbines at its Colossus data center near Memphis without air permits. The court rejected xAI's argument that the turbines are exempt as mobile sources, as well as its constitutional challenge to citizen suits. The ruling lets one of the most prominent environmental cases against AI data centers move forward. A similar motion to dismiss from the Justice Department was denied earlier in October.
Washington Post: Microsoft Copilot Leaned on an Error-Prone, AI-Built Voter Guide Days Before the Midterms
A Washington Post investigation found that Microsoft Copilot cited the Ballot Brief, a largely AI-generated voter-guide site, in 75% of the Post's voter-guide tests, compared with 22% for ChatGPT. The Post found the site matched ZIP codes to the wrong congressional districts, and chatbots passed those errors on to users. With 2 in 5 Americans saying they have used AI to look up election information, it shows how one low-quality AI source can spread through chatbot answers at scale.
Previous digests
Oct 31The iShares Semiconductor ETF ended the week 29% below its recent high, which puts it in a...6 stories
Oct 30OpenAI published a detailed account of the four-layer system it now uses to watch its fron...6 stories
Oct 29Anthropic's Frontier Red Team says an unreleased internal research model found flaws in fi...5 stories
Oct 28Google DeepMind unveiled Genie 4, an interactive world model that generates native audio (...6 stories
Oct 27Anthropic filed an S-1 with the SEC on October 26 to list on Nasdaq under the ticker ANT. ...6 stories