Google's Threat Intelligence Group says it caught the first known case of a real attacker using a large language model to find and weaponize a zero-day - a 2FA bypass in a popular but unnamed open-source web-based system administration tool. Google has high confidence the Python exploit was AI-generated, citing textbook code structure, abundant educational docstrings, and a hallucinated CVSS score in the script. The flaw was a high-level logic bug, the kind LLMs excel at spotting, rather than a memory corruption issue. Google rules out Gemini and warns that AI-assisted exploit development is being industrialized via account-pooling and proxy relays for premium models.
SailPoint, the identity governance vendor used by many large enterprises, disclosed in a SEC 8-K filing that attackers gained unauthorized access to a subset of its GitHub repositories on April 20. The company's incident response team contained the intrusion the same day. SailPoint says no customer data in production or staging was accessed and its services were not interrupted. The root cause was a vulnerability in a third-party application, which has been remediated. SailPoint notified affected customers directly and says no further customer action is needed. The company has not disclosed what data was actually in the impacted repos.
The UK Information Commissioner fined South Staffordshire Water 963,900 pounds over a 2022 Cl0p ransomware breach that exposed 633,887 customer and employee records. The penalty notice reveals attackers were inside the network nearly two years before discovery - initial access happened September 2020 via a malicious email attachment, but they were not detected until July 2022 when IT performance issues triggered an investigation. The ICO found basic security failures: an unpatched ZeroLogon flaw on two domain controllers, no principle of least privilege, an outsourced SOC monitoring just 5 percent of the IT estate, and Windows Server 2003 boxes still running in production.
Instructure confirms that ShinyHunters exploited multiple cross-site scripting flaws in Canvas to deface school login portals on May 7, demanding the company and individual schools negotiate ransom by May 12. The flaws are in user-generated-content features of the free Free-for-Teacher Canvas environment and let the attacker grab authenticated admin sessions. This was a second hit following the original breach disclosed a week earlier that ShinyHunters claims netted 3.6 terabytes covering 8,809 educational organizations and 275 million student, teacher, and staff records. Instructure has taken Free-for-Teacher offline and applied additional safeguards; main Canvas has been restored since May 9.
The TrickMo Android banking malware now routes its command-and-control through The Open Network (TON), the decentralized peer-to-peer network originally built around Telegram, making the C2 infrastructure much harder to identify or take down. ThreatFabric (which tracks this variant as Trickmo.C) has been watching it since January in campaigns hitting users in France, Italy, and Austria. The malware disguises itself as TikTok or streaming apps and steals banking credentials and crypto wallet keys via phishing overlays, keylogging, SMS interception, OTP suppression, and live screen recording. The new variant also adds SSH tunneling, port forwarding, and SOCKS5 proxy commands, turning infected phones into a pivot point.
A researcher at Israel Aerospace Industries published a proof-of-concept tool called GhostLock that uses a legitimate Windows API call to make files unreadable without encrypting anything. The technique abuses the dwShareMode parameter of CreateFileW - setting it to 0 grants the calling process exclusive access, so every other user or app trying to open the file gets a sharing violation. GhostLock automates this recursively across SMB shares from a standard domain user account, no elevation required. Researcher Kim Dvash frames it as a disruption attack, not destructive - data is not lost, but operational downtime can mirror a ransomware incident.
QiAnXin XLab has tied the ongoing exploitation of cPanel's CVE-2026-41940 to a previously-quiet threat actor it tracks as Mr_Rot13, who has been operating since at least 2020. The attack chain exploits the cPanel and WHM authentication bypass to drop a Go-based infector that adds an attacker SSH key, plants a PHP web shell, and serves a fake login page to steal cPanel credentials (ROT13-encoded, exfiltrated to wrned[.]com). The final payload is a cross-platform backdoor called Filemanager that runs on Windows, macOS, and Linux. XLab counts over 2,000 attacker source IPs currently scanning for this flaw.
Researchers at Cyera disclosed a critical bug in Ollama, the open-source tool that runs large language models locally on laptops and servers. The flaw, called Bleeding Llama (CVE-2026-7482), lets anyone with network access send a malformed model file and read raw process memory back - which typically contains API keys, environment variables, system prompts, and other users' chat history. Ollama ships without authentication by default, so an estimated 300,000 instances are exposed on the internet. Ollama 0.17.1 fixes it. Separately, Striga disclosed two unpatched Ollama Windows desktop flaws (CVE-2026-42248 and CVE-2026-42249) that chain into persistent code execution at login.
Hackers are buying Google ads that look like they go to claude.ai - and they do go to a real claude.ai page. But the page is a shared Claude chat dressed up as 'Apple Support' walking users through installing Claude on a Mac. The instructions tell people to paste a command into Terminal that quietly downloads MacSync, a Mac infostealer that grabs saved browser passwords, cookies, and contents of macOS Keychain (where Mac stores logins and keys). Because both the ad and the page are real claude.ai links, there is no fake domain to spot. Researcher Berk Albayrak first reported the campaign; BleepingComputer found a second active variant.
Zara is the latest big brand caught in the ShinyHunters extortion campaign tied to the March breach of analytics provider Anodot. The attackers - who got into Anodot in March and used that foothold to raid Snowflake-hosted data for at least a dozen downstream customers - have now published roughly one terabyte of files they say came from Zara's customer support system. Have I Been Pwned loaded 197,376 unique email addresses from the dump, along with product SKUs, order IDs, and the market each support ticket originated in. Zara's parent Inditex says no passwords or payment data were exposed.
TrustStrike Labs