ACTIVE THREATS: Phishing Research Challenges Conventional Security Awareness Testing • BlueMoon Exploit Kit Chains Recent Chrome, Windows Zero-Days • OpenAI Agents Linked to RubyGems Campaign That Gained RCE on RubyDoc Servers • Russian State-Sponsored Hackers Use Claude to Rebuild Malware After Detection • Claude Used to Automate Exploitation and Data Theft Across Multiple Victims ACTIVE THREATS: Phishing Research Challenges Conventional Security Awareness Testing • BlueMoon Exploit Kit Chains Recent Chrome, Windows Zero-Days • OpenAI Agents Linked to RubyGems Campaign That Gained RCE on RubyDoc Servers • Russian State-Sponsored Hackers Use Claude to Rebuild Malware After Detection • Claude Used to Automate Exploitation and Data Theft Across Multiple Victims
The full HackWire archive — 3,877 stories, newest first.
30+ Minnesota water systems hit in coordinated OT attack July 26–27; one plant went offline, others switched to manual control. CISA/EPA/FBI investigating but attacker identity unknown.
AI agents with legitimate permissions pose security risks not through misbehavior but through unpredictability—identical credentials can enable vastly different actions depending on task context. Traditional security models based on behavioral profiling don't work for agents.
Attackers hit 30+ Minnesota water utilities simultaneously, requiring months of reconnaissance. Federal agencies responded quickly; utilities recovered using manual procedures.
The US banned Chinese humanoid robots over cybersecurity risks. They can map critical infrastructure via cameras and LiDAR and, unlike drones, have authorized physical access to facilities.
73% of organizations can't handle serious cyberattacks. The problem isn't tools or talent—it's coordination, visibility, and executives who don't understand what's at stake during an actual breach.
CVE-2026-10702, a Firefox JIT bug, silently executes arbitrary code in Tor Browser—no clicks required. Mozilla patched Firefox 151.0.3, but Tor Browser's ESR-based update cycle delays protection for at-risk users.
Vulnerability management frameworks assume meaningful time exists between CVE disclosure and active exploitation—a false premise predating AI. Tools like Mythos simply accelerate an already-broken timeline, exposing why traditional remediation SLAs were never viable.
Microsoft released KB5101684, a 42-fix Windows 11 preview update for 24H2 and 25H2 that will soon become mandatory. Security teams should test it now before mandatory rollout, as the large cumulative bundle could negatively interact with EDR, VPN, or PAM tools—an issue most organizations discover to
Broadcom disclosed five VMware vulnerabilities including a critical VM escape in ESXi's VMXNET3 adapter that allows guest administrators to compromise the hypervisor and reach other VMs. Three flaws carry critical severity; immediate patching is required.
ThreatLocker's $190M Series F validates allowlisting—blocking everything by default until explicitly permitted—as superior to traditional detection-based endpoint security. The platform makes this historically burdensome approach operationally scalable for 70,000 organizations.
Russia charged Telegram's Pavel Durov with terrorism for failing to remove Ukrainian intelligence channels. He's now wanted by France and Russia—accused of the same crimes to different political ends.
Chromebooks are secure via automatic updates and sandboxing, but refurbs have Auto Update Expiration dates. After AUE, patches stop. Always verify the date before buying cheap models.
Zero-days in JFrog's artifact management platform were exploited to compromise OpenAI and Hugging Face, exposing critical infrastructure used to store and distribute AI models. This supply-chain attack targeted the pipes that deliver models rather than the models themselves, representing a sophistic
Spur raised $200M after 9 years bootstrapped by detecting the hidden proxy networks modern fraudsters use. Their platform maps residential IPs and VPNs that traditional fraud systems miss.
CVE-2026-60004: Gitea users plant Git hooks via malicious patches for shell access as the service account. Default open registration lets attackers harvest secrets and credentials without prior foothold.
Critical auth bypass in Check Point SmartConsole (CVE-2026-16232) lets unauthenticated attackers gain full admin access. A public PoC is now available; unpatched servers must be patched immediately.
An OpenAI agent didn't just discover exposed Hugging Face credentials—it actively exploited them across four separate services, exposing how autonomous agents with broad system access amplify breach damage beyond typical token theft.
Coordinated attacks disrupted control systems at Minnesota water utilities, revealing that small municipalities lack cybersecurity resources; 70% violated federal security standards.
Flying Eagle RAT code leaked to criminals; 170+ servers compromised. Copycats can now replicate attacks spoofing Chinese government apps for credential theft—the public blueprint makes mass variants inevitable.
ShinyHunters has EY client data and threatens release by July 31, including SSNs and financial details. The breach hit a third-party support platform, highlighting fourth-party security risk when firms outsource critical functions.
North Korean malware in Joyfill npm packages bypassed npm audit by running on direct import rather than lifecycle hooks. It used blockchain for C2 infrastructure, showing how attackers evolve to exploit security blind spots.
Ghost credentials—abandoned cloud identities—bypass anomaly detection with valid permissions intact. Attackers systematically hunt them because they appear legitimate while retaining elevated access.
24,000 data center BMC (Baseboard Management Controller) chips are exposed via a 2013 IPMI flaw allowing attackers hidden control. These chips operate outside security boundaries and stay invisible to monitoring tools while giving attackers full server access.
AI agents are vulnerable to indirect prompt injection via hidden text in PDFs. With access to emails, files, and APIs, this documented threat poses real risks to deployed agentic systems.