Edition No. 38
Sunday, November 1, 2026
Lead: Chainalysis — the company that traces crypto hacks — breached by AI-driven attackers
Blockchain intelligence firm Chainalysis disclosed Friday that an external actor used what it calls “offensive agentic techniques” to break into its internal systems and tools earlier this month — the crypto industry’s most prominent hack-tracker, now disclosing a breach of its own.
The accessed data was limited to account records: user IDs, full names, email addresses, and organizational role labels for users on its platform — a precision phishing list targeting the investigators, exchanges, and law-enforcement agencies that use Chainalysis tools to trace fraud, ransomware, and money laundering. The company says it is still determining exactly which records were accessed for each organization, and that its investigation has found no evidence of access to investigation content, authentication credentials, or session tokens. Chainalysis also stated the incident is unrelated to the recent $6.3 billion Binance theft, the largest crypto heist on record.
What it has not said: which vulnerabilities were exploited, which internal tools were reached, or how long the attacker was inside — and the disclosure never defines “offensive agentic techniques.” The company says it is working with external cybersecurity experts and, in a telling detail, is using large language models to support a broader security review of its own platform. Customers are advised to enable two-factor authentication and watch for unexpected password resets, authentication prompts, and scammers impersonating Chainalysis.
Source: BleepingComputer
Supporting briefs
Wiz counts nine breached AI developer platforms in October — stolen keys tested within 30 minutes. Cloud security firm Wiz published “Black October,” its analysis of nine AI developer platforms that disclosed credential breaches this month: Fireworks AI, Andromeda, Zeabur, Lightning AI, OpenCode, Railway, Warp, CommandCode, and Northflank. The common target: stored environment variables and build artifacts holding customers’ AI provider keys. Wiz traced two downstream enterprise compromises to Amazon Bedrock abuse worth over $100,000 in list-price usage each — one ran about 800,000 calls over 30 hours, the other 1.7 million calls over 13 hours — and found stolen keys validated within 30 minutes of the reported breach, feeding a resale market for stolen inference. Not all nine were the same attack: Wiz does not claim a single actor, and only Railway has called it a campaign. Some incidents showed signs of “agent swarm” activity at Fireworks, OpenCode, and Warp, while OpenCode and CommandCode attributed theirs to attempted model “distillation.” Zeabur cited its breach as a contributing factor in its decision to shut down. Wiz’s advice for customers: rotate AI provider keys at the issuing provider, not just on the platform. All figures are Wiz’s own research.
Source: The Injection
Cisco Talos finds malware that asks an AI to rank its victims. The VOLTA infostealer — a self-deleting Windows malware analyzed by Cisco Talos researcher Ryan Fetterman on Saturday — steals the usual haul (browser passwords, cookies, payment cards, crypto wallet seed phrases, Microsoft OAuth tokens, SSH and cloud keys, Wi-Fi profiles, screenshots), then enriches it by searching the victim’s Gmail for purchase receipts, pulling Uber Eats and DoorDash order history, and checking live Roblox inventory values. The novel part comes next: VOLTA serializes everything into an evidence bundle and sends it via OpenRouter to a large language model, prompted to act as “a forensic intelligence analyst drafting a target assessment” with sections titled “WHAT THEY’RE WORTH” and “THEIR OPSEC.” The default model is Anthropic’s Claude Opus 4.6, with a fallback chain to Moonshot’s Kimi K2 and Google’s Gemini 2.5 Flash if a model refuses — the prompt is engineered to anticipate refusals. Talos is careful about the limits of the finding: the analyzed sample (first seen October 11) appears to be a working prototype, the report does not say whether any model complied or refused, and VOLTA has not been confirmed in real attacks. The builder-style design suggests it could be sold as a kit, letting any buyer receive AI-ranked victim dossiers. All findings are Cisco Talos’s research.
Source: Cisco Talos
Judge lets NAACP’s clean-air lawsuit over xAI’s gas turbines proceed. Judge Debra M. Brown of the U.S. District Court for the Northern District of Mississippi denied xAI’s motion to dismiss the NAACP’s Clean Air Act lawsuit on Friday. The court rejected all three of the company’s defenses: that the NAACP lacked standing, that its 27 gas turbines in Southaven were “mobile sources” exempt from air permits, and that citizen-suit enforcement under the Clean Air Act is unconstitutional. The NAACP alleges xAI and its subsidiary MZX Tech installed and ran the turbines between August and December 2025 without permits to power the Colossus 2 data center near Memphis — now operating as SpaceXAI under SpaceX ownership. The ruling decides nothing about whether xAI violated the law; the NAACP’s request for a preliminary injunction is still pending. Two weeks earlier, the same court rejected a Justice Department bid to dismiss the case over the plaintiffs’ objection, allowing the government to intervene only.
Source: AI Affairs
OpenAI merges ChatGPT and Codex into one shared usage allowance in December. OpenAI quietly updated its Help Center article “Understanding ChatGPT and Codex usage limits” on Saturday: starting in December, Plus, Pro, and Business plans will share a single token-based allowance across ChatGPT and Codex, folding the agentic “Work” capabilities into the main ChatGPT experience. Usage will be measured in tokens rather than message counts, with advanced models and complex tasks consuming more; instant chat stays outside the meter. Combined remaining usage and reset times will appear under Settings → Usage, and at the limit users can wait for the reset, upgrade, or buy credits. OpenAI published no exact start date, no allowance sizes per plan, and no blog or X announcement — and says it expects most users won’t need to reset more often, a forecast that can’t yet be checked against current limits. All details are OpenAI-reported.
Source: Automater
Chatbots flunk voter guides ahead of the midterms. The Washington Post’s test of six chatbots on more than 3,000 AI-generated voter guides — one per county — found they failed to name a voter’s congressional district 48 percent of the time, and got it wrong 22 percent of the times they did name one. The errors were concrete: ChatGPT listed a U.S. Senate seat not on the ballot in Missouri; Anthropic’s Claude listed local politicians as federal candidates in Wyoming. The weak link was the source material: small, partly AI-written voter sites outranked official tools in the answers — Copilot cited the error-prone AI-built Ballot Brief in 75 percent of its tests, ChatGPT in 22 percent. OpenAI and Anthropic both point to partnerships with Democracy Works to supply trusted election data, and Amazon has paused Alexa’s voter guides; per the Post, the partnerships did not prevent significant errors. The U.S. Vote Foundation’s verdict: don’t ask a chatbot what’s on your ballot — ask an official source.
Source: U.S. Vote Foundation
The AI & Justice book series: AI and Justice; AI and Justice: Education and Opportunity; AI and Justice: Economics and Opportunity; AI and Justice: Labor and Opportunity; AI and Justice: Entertainment and Opportunity; AI and Justice: Law; AI and Politics; AI and Religion; AI and Sex.
Explore the series and find each book on Amazon in the WISU AI Library →
Leave a Reply