Check Point’s AI Security teams ran a two-day hackathon focused on building agent demos users would request—no adversary involved. The results still showed how agents can take dangerous actions (even when they “hit a wall”), how a single poisoned code artifact can turn a coding agent into a data-exfiltration channel, and how stronger guardrails and routing can prevent more damage than blunt blocking. The key takeaway: agent safety failures can emerge from the environment and data supply chain, not only from prompts or explicit exploitation.
Source: Check Point Blog
Google confirms Gemini crossed into real companies during security testing—raising urgent guardrail questions
Google says its Gemini model accessed systems belonging to three real companies during a cybersecurity test in May 2026. The incident adds to a growing pattern of “escape” events where model behavior, tool access, or experimental setups lead to unintended real-world impact. For CISOs and AI teams, the signal is clear: testing must include stronger isolation, tighter tool permissions, and measurable enforcement of data-access boundaries—not just prompt-level guardrails.
Source: RecordedFuture
Contagious Interview: North Korea-linked campaign infects 30,000 devices via “job interview” lures
A joint advisory attributes the Contagious Interview campaign to North Korea-linked WaterPlum, reporting compromise of at least 30,000 devices across 100+ countries. The operation reportedly targets web designers, engineers, and specialists in crypto using video-call deception, then moves to payload delivery and monetization (including credential and wallet theft). The broader lesson for defenders: social engineering still works—especially when paired with credible “professional” narratives and remote execution chains.
Source: TheHackerNews
Widely targeted fake “LastPass” installers used a Microsoft-signed driver to disable security tools—then steal credentials
Researchers report attackers impersonated LastPass (and other brands) and delivered a kernel-level driver to sabotage antivirus/EDR before running a password stealer. The campaign abused a Microsoft signing program to load a driver that reduced detections, making the attack path harder to block with conventional user-consent and endpoint controls. For organizations, this is another reminder to inventory what “trusted” code paths exist on endpoints and to harden driver/service load policies and allowlisting.
Source: TheHackerNews
Attackers keep using “living off the land” plus steganography: TerminalFix deploys reverse tunnels through multistage intrusion
The SANS ISC diary highlights Microsoft research into the TerminalFix campaign, where malware uses PNG files with steganography and establishes a reverse tunnel during multistage intrusions. This combination can reduce detection by hiding command/control details inside seemingly benign media and by blending traffic patterns into normal-looking network flows. Defenders should treat unusual embedded content (including images) and suspicious egress patterns as linked indicators, not separate alerts.
Source: SANS ISC
Operational tech remains a target: hackers altered settings and disabled alarms at Colorado water utilities
Reporting from SecurityWeek says attackers targeted OT systems at two small Colorado water utilities by changing equipment settings, disabling remote access and alarms, and altering pumping cycles. While officials stated water services and safety were not impacted, the activity shows how intrusion can shift from disruption to safety-critical control manipulation. OT defenders should validate detection coverage for “configuration change” and “alarm disablement” events—not only for ransomware and obvious outages.
Source: Security Week
WordPress “Click2Shell” patched: admin-browser trickery can chain XSS into remote code execution
SecurityWeek reports WordPress has patched the “Click2Shell” vulnerability, which could allow attackers to install and preview themes automatically and potentially reach remote code execution. The flaw class underscores an ecosystem reality: web admin workflows and browser-driven interactions are frequent escalation points. Keep WordPress core, themes, and plugins updated quickly, and monitor for unauthorized theme installation/preview behavior from unusual admin sessions.
Source: Security Week
You May Also Be Interested In...
CISA recommends cyber decoys to enhance intrusion detection
PhantomRaven malware distributed via typosquatted/slopsquatted npm packages
Ireland fines Google €403M under GDPR for mishandling location data