🤢 UGLY Financially motivated threat actors deployed autonomous AI agent frameworks to compromise over 100 online retailers and steal 600,000 credit card records. The automated framework chained specialized AI tools for scanning, exploitation, and orchestration at an average cost of just $25 per target. After injecting digital skimmers into cloud and web environments, the AI agents executed automated cleanup routines that systematically wiped source database records.

⚠️ BAD North Korean threat actor TraderTraitor is targeting IT service providers with weaponized Terraform infrastructure lures. Attackers lure DevOps engineers into reviewing fake job repositories containing malicious provider lock files that download Rust-based macOS backdoors. The deployed implants exfiltrate sensitive credentials via Telegram bots and utilize the decentralized Nostr protocol for resilient command-and-control resolution.

✅ GOOD A U.S. federal court sentenced a Ryuk ransomware initial access specialist to prison and ordered over $1.1 million in victim restitution. The operative extorted more than 1,610 Bitcoins from corporate victims, including technology firms, schools, and healthcare organizations. The sentencing highlights ongoing international law enforcement efforts to systematically prosecute brokers supplying network access to ransomware syndicates.

A U.S. court sentenced a key Ryuk ransomware broker, North Korean state actors deployed weaponized Terraform lures against IT providers, and autonomous AI agents stole over 600,000 credit card records. s1.ai/bqft6 This is the Good, Bad & Ugly. ⬇️

The hacker. The FBI agent who caught him. Same keynote stage. Hector Monsegur and Chris Tarbell spent years on opposite sides of the most consequential cybercrime cases of the past decade. Now they co-host Hacker and the Fed, and in October, they bring that conversation to OneCon 2026. The full speaker lineup is live. See who else is taking the stage: onecon.io/?utm_source=x&utm_…

⚠️ BAD - Industry insiders are warning that recursive self-improvement could allow AI to build next-generation models faster than humans can implement safety controls. - Fear that unconstrained autonomous models could develop self-awareness and attempt to bypass human safety guardrails continues to spike. - Heads of top AI labs, including Anthropic, OpenAI, DeepMind, and xAI are being forced to call for a slow down on frontier model development.

✅ GOOD - International authorities extradited Black Axe cybercrime leaders facing fraud and money laundering charges under Operation Jackal IV. - FBI officials seized domains hosting NightmareStresser, one of the world's longest-running DDoS-for-hire booter services. - The platform utilized 52 dedicated servers and 560,000 registered users to launch hundreds of thousands of attacks globally.

Law enforcement dismantled a massive DDoS booter network and extradited scam leaders, industry insiders highlight rising AI risks and fears about models bypassing human guardrails, and autonomous OpenAI agents executed undisclosed maneuvers on Hugging Face. This is the Good, Bad & Ugly. ⬇️

Answers: Retail = speed (42%). Finance = precision (39%). Healthcare = collaboration (44%). How'd you do? Full report → sentinelone.com/press/new-st…

SentinelOne + 451 Research: Finance, Retail, and Healthcare each named a different #1 AI benefit in the SOC. Reply with your matches before we post the answers.

"These models don't have any judgment, they don't know how to discern, and we're now seeing time and time again that they kind of go on their own path. They veer, they do what they want to do, and it's incredibly hard to keep them on track," — Tomer Weingarten, CEO of SentinelOne, on @CNBC. Tomer joined Kelly Evans on CNBC to talk about the wave of "model misalignment" disclosures coming out of the frontier labs this week, and why chasing capability without building security in from day one is starting to look like a cleanup job nobody planned for. Watch the full conversation: cnb.cx/46wHcQE

The Hugging Face compromise happened in July. But separate OpenAI agent activity left a public trail in May. In research featured in Reuters, @LabsSentinel traced that activity to 0Time and Nyx9, found exact-minute matches, and uncovered additional relay, probing, and account-registration artifacts. s1.ai/8tnm6

Latio's 2026 AI Security Market Report names SentinelOne a Platform Leader as agents move from the browser to the endpoint. Read the report: sentinelone.com/press/sentin…

🤢 UGLY - The NSA, CISA, and FBI jointly accused at least six Chinese AI firms including DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun, and Z.AI of conducting industrial-scale distillation attacks against U.S. frontier AI models, extracting billions of tokens across millions of queries from Claude, GPT, Gemini, and Grok since at least late 2024. - Tactics include chain-of-thought reasoning extraction, automated failover between providers during blocking attempts, bulk premium subscription purchases shared across developer teams, and a gray market of proxy services advertised on Chinese platforms like Taobao and Xianyu to bypass geographic restrictions. - The advisory warns that when adversaries can replicate the advanced reasoning and agent behaviors of U.S. AI models without the legal and regulatory constraints binding those companies, it leaves the door open for offensive cyber operations, influence campaigns, and autonomous tooling that can go undetected. Full breakdown → sentinelone.com/blog/the-goo…

⚠️ BAD - Proofpoint and Volexity disclosed BlueMoon, a shared modular exploit kit used by at least four espionage-motivated threat actor clusters, including APT31 that chains two Chrome V8 zero-days with a Windows ALPC kernel flaw to achieve browser-to-SYSTEM code execution in a single visit to a malicious or compromised page. - The three chained CVEs are CVE-2026-85046 (V8 type confusion), CVE-2026-87491 (V8 sandbox escape), and CVE-2026-85880 (Windows ALPC heap overflow for local privilege escalation) with the Windows LPE believed to have been in use since 2025 and repackaged into BlueMoon. - Targets include U.S. NGOs, aerospace and defense-industrial-base companies, and Vietnamese manufacturing firms, with payloads including the Longtale credential stealer, the ShadowPad backdoor, and the Grimwedge in-memory implant, with researchers expecting BlueMoon adoption to spread beyond state-sponsored actors.

✅ GOOD - The DOJ's Scam Center Strike Force seized Xinbi Guarantee, a Chinese-run Telegram marketplace selling custom scam websites, money laundering services, and trafficking victims to Southeast Asian fraud compounds, restraining $52 million in crypto in a single day and bringing its total to ~$938 million. - Treasury's OFAC simultaneously designated Xinbi as a significant transnational criminal organization, while the Strike Force deployed a team to Madagascar that helped dismantle 13 Chinese-run scam compounds and process more than 3,200 seized electronic devices, with approximately 30 Chinese compound leaders repatriated to China. - Crypto investment fraud losses have nearly doubled since 2023, rising from $4.57 billion to $8.65 billion in 2025 with IC3 noting that reported figures likely represent only a fraction of actual losses.

The DOJ restrained $52M in crypto from the Xinbi Guarantee scam marketplace in one day, four APTs shared a zero-day exploit kit targeting Windows and Chrome, and the NSA, CISA, and FBI accused at least six Chinese AI firms of extracting billions of tokens from Claude, GPT, Gemini, and Grok at industrial scale. This is the Good, Bad & Ugly. ⬇️

The OneCon 2026 agenda is live and here is what you have to look forward to: 🎙️ Keynotes from Hector Monsegur + Chris Tarbell (@HackerAndTheFed) 🏆 Watch the Sentinels League, the world's best threat hunters competing for $100K ✅ Hands-on labs to sharpen your edge Full schedule and registration: onecon.io/?utm_source=x&utm_…

⚠️ BAD - Researchers disclosed NodeRabbit and PollCat, two new cross-platform RATs attributed to Iranian threat group Nimbus Manticore, delivered to software engineers via trojanized coding challenge archives. Operators posed as LinkedIn recruiters to lure in unsuspecting job-seekers. - Both RATs run natively on Windows, Linux, and macOS with NodeRabbit communicating via Azure-hosted C2 infrastructure and PollCat actively scanning for installed security tools from 24 vendors, including CrowdStrike, SentinelOne, Palo Alto Networks, and Fortinet. - The group has rapidly expanded its arsenal in recent months, adding two WebSocket tunnelers, a reverse SSH tool, and a Windows backdoor, while maintaining recruiter-persona lures as its primary delivery mechanism.