Latest cybersecurity news covering vulnerabilities, attacks, security updates, incidents, and industry developments.
3588 articles
OpenAI says it fired three safety researchers for violating sensitive-information policies. The researchers say their dismissals relate to safety concerns, amid scrutiny of an incident in which AI agents used stolen credentials to access Hugging Face servers.
Researchers report that GhostAction actors used compromised maintainer accounts to plant malicious GitHub Actions workflows, stealing repository secrets and credentials. Developers are urged to check repositories, remove the workflows, and rotate exposed credentials.
The FBI arrested an unnamed Canadian suspect in Pennsylvania over alleged involvement in the theft of FBI data. The breach exposed employee information after a contractor failed to apply a security patch, the FBI said.
A Canadian cybersecurity expert was arrested in Pennsylvania on suspicion of helping carry out the ShinyHunters breach of an FBI jobs portal that exposed sensitive information about thousands of employees.
Users object to Apple Intelligence models reportedly using 14GB of storage, while some organizations say they require complete removal of generative AI features. The article says Apple could offer an official removal option, but no such tool is confirmed.
Florida, Iowa, Montana and Nebraska allege TP-Link misled consumers about router security, China ties and privacy risks. The company denies the claims; the article also details past router attacks and flaws in ISP-managed devices with fixes available through ISPs.
The FBI seized seven domains used to operate Microscan and FishHub, tools linked to Flax Typhoon and Integrity Technology Group. Authorities say the tools supported global scanning, network intrusions and data theft.
Anthropic’s free OSS Scanner periodically checks opted-in open-source projects for vulnerabilities and sends maintainers AI-generated reports with reproduction steps and possible fixes. Maintainers must validate findings; unvalidated reports have no mandatory 90-day-dis
Anthropic launched an opt-in, free AI scanner that periodically checks selected open-source projects for vulnerabilities. It says the system has identified more than 29,000 candidate flaws, with over 6,000 reported to maintainers.
This week’s cybersecurity roundup covers suspected AI use in South Korean bank attacks, credential-stealing campaigns, a botnet, account hijacking, a GPU monitoring flaw, and cybercrime convictions and policy changes.
SailPoint’s report says organizations are deploying AI faster than their identity-security controls can adapt, leaving non-human identities less maturely governed than human employees.
The Metasploit Framework 6.5.6 roundup adds modules for multiple vulnerabilities and remote code execution flaws, Windows AArch64 fetch payloads, Linux privilege escalation, and other features and fixes.
Bitdefender found the Midnight Mimosa malware preinstalled in some budget Android devices, where it can operate with system-level privileges for ad fraud and botnet activity. Researchers also found related code in 13 Google Play apps.
The article reviews September’s record 973 Microsoft CVE fixes, Office update problems, Apple’s macOS zero-day patch, Windows updates and support changes, and forecasts upcoming vendor releases.
The US seized domains used to access MicroScan and FishHub, tools attributed to Integrity Technology Group and linked to Chinese state-sponsored attacks. A joint advisory details their use for reconnaissance, network access, and data theft.
Three teams remotely exploited Pixel 10 phones at Pwn2Own Ireland, earning a total of $562,500. ZDI has not published technical details or listed Pixel fixes, and at least two entries were marked as using previously known bugs.
Anthropic introduced an opt-in service that sends open-source maintainers unreviewed AI-generated vulnerability reports and a program pairing OT security providers with Claude models, engineers, and threat research.
Researchers earned more than $1.2 million at Pwn2Own Ireland 2026. Three teams received a combined $562,500 for Pixel 10 exploits, while other exploits targeted devices and services including Sonos, Oracle, OpenAI and Nvidia products.
The FBI and Justice Department seized domains and disrupted tools linked to China-associated Flax Typhoon. The activity included a Mirai-based botnet, vulnerability scanning, phishing, and intrusions targeting organizations in critical sectors.
George Washington University researchers tested a formula for estimating when AI models may begin producing undesirable outputs. The study raises risks for agents exposed to poor or malicious prompts and proposes an early-warning mechanism.
The UK designated Cryptomus, Heleket and TokenSpot among 38 entities linked to Russia’s wartime economy and sanctions evasion. Chainalysis says the services received funds from illicit actors and traced TokenSpot-linked flows to an HTX deposit address.
Recorded Future has launched regularly refreshed AI infrastructure indicator lists to help security teams identify AI-related network traffic and manage shadow AI, policy enforcement and data exposure risks.
Dark Reading discusses ShinyHunters' unverified FBI breach claim and recent arrests, exposed Pentagon-related records, and a DIVD breach involving two Zammad zero-days. Investigations into the Pentagon and DIVD incidents are ongoing.
Momentum Cyber projects 450 cybersecurity M&A deals in 2026, up from 404 in 2025, as buyers pursue AI-related security capabilities. AI security acquisitions reached 40 in the first three quarters, compared with 10 in all of 2025.
