Latest stories — Page 40
Treat the Model Like a Threat: Why AI Agent Security Needs a Systems Overhaul
A paper from researchers at Google and two US universities argues that prompt-level defences and alignment tuning are structurally inadequate for securing autonomous AI agents — and that enterprises should start treating the model itself as an untrusted component.
Ten Thousand Bugs, One Model: Inside Anthropic's Project Glasswing
Claude Mythos Preview has scanned more than a thousand open-source projects and surfaced thousands of critical flaws. The bottleneck has moved — and the patch queue is not moving fast enough.
Kali365 and EvilTokens: The New Phishing-as-a-Service Threat
Professional phishing kits lower barriers for attackers, bypassing MFA with ease.
Unpatched Flaws Now Outpace Stolen Credentials as the Leading Breach Entry Point
Verizon's 2025 DBIR puts vulnerability exploitation at 31% of breach root causes. Median patch time has climbed to 43 days, and only 26% of CISA KEVs were fully remediated — a gap attackers are sprinting through.
AI Governance Is Broken Because It Still Lives Outside the Pipeline
Building compliance as a post-ship review layer made sense for static software. For AI systems that mutate overnight, it is organizational negligence dressed up as process.
TrapDoor: The Supply Chain Campaign That Wants Your Whole Dev Environment, Not Just Your Secrets
A cross-registry malware campaign hitting npm, PyPI, and Crates.io is going after CI/CD pipelines, SSH trust chains, and AI coding assistant files — not just credentials on install.
Twenty Years of Cyber Lessons and We're Still Losing on the Basics
The industry spent two decades reinventing its philosophy — perimeter defense to assume-breach — yet the attacks that still hit hardest exploit the same unpatched, misconfigured, un-MFA'd mistakes we should have buried years ago.
AI Agent Identities Are Redrawing Enterprise IAM Budgets
New Omdia research finds that the rapid spread of AI agent deployments is forcing organisations to treat non-human identities as a distinct governance category, with budget implications that traditional identity and access management frameworks were not designed to absorb.
Two Defender flaws under active exploitation, Microsoft confirms
A SYSTEM-level link-following bug and a denial-of-service issue in Microsoft Defender are both being abused in the wild.
The Boring Attacks Are Winning: Why Defenders Keep Losing to Trusted Tools
Leaked tokens, poisoned npm packages, and login replays are doing more damage than zero-days this quarter. Here is how to spot the pattern before it spots you.
Showboat: A Modular Linux Backdoor Quietly Camped in a Middle East Telco Since 2022
Lumen's Black Lotus Labs ties the SOCKS5-capable implant to a years-long intrusion at a regional carrier, with an in-memory loader and ELF payloads that sidestep most host telemetry.
Cisco's Secure Workload Earns a Perfect 10, in the Wrong Sense
An unauthenticated REST API flaw rated CVSS 10.0 lets remote attackers help themselves to sensitive data. Cisco has issued fixes.
CISA's KEV List Just Picked Up Langflow and Apex One — Both Already Being Hit
Two flaws, one AI workflow tool and one veteran endpoint suite, now carry a federal patch deadline because attackers got there first.
When the Hardware Isn't There: Coaxing Vulnerable Drivers Into Range
BYOVD research keeps colliding with a stubborn problem — many kernel drivers refuse to talk unless their device is plugged in. New work shows how to make them talk anyway.
Justice Department Charges Ottawa Man With Operating Kimwolf DDoS Botnet
Federal prosecutors say Jacob Butler, 23, developed and rented out a variant of the AISURU botnet for paid denial-of-service attacks.
Megalodon Campaign Plants Malicious Workflows in 5,561 GitHub Repos in Six Hours
Throwaway accounts pushed 5,718 commits forging build-bot identities to exfiltrate CI/CD secrets, researchers said.
CERT-UA Attributes Prometheus-Themed Phishing Run Against Ukrainian Government to Ghostwriter (UAC-0057)
Compromised mailboxes deliver lures impersonating a Ukrainian e-learning platform, with the Belarus-aligned operator tracked as UNC1151 named as the responsible cluster.
Operation Saffron Yanks the Plug on First VPN, the Getaway Car of at Least 25 Ransomware Crews
French and Dutch police led the takedown of a bulletproof VPN service that prosecutors say routed traffic for Conti, LockBit affiliates, and roughly two dozen other ransomware brands.
CISA Flags Exploited Drupal SQL Injection Flaw. Drupal Won't Say Who Got Hit.
CVE-2026-9082 is in the Known Exploited Vulnerabilities catalog. The advisory mentions active exploitation. It does not mention victims, telemetry, or how anyone found out.
LiteSpeed cPanel Plugin Flaw Hands Root to Any Logged-In User, and the Vendor Won't Say How Many Hosts Are Hit
CVE-2026-48172 carries a CVSS of 10.0, is already being exploited, and LiteSpeed has not answered three questions about exploitation telemetry.
Laravel-Lang Packages Hijacked to Push a Cross-Platform Credential Stealer
Four popular Laravel-Lang packages were tagged with malicious releases that drop a credential-harvesting framework on Windows, macOS, and Linux.