This Week in Cybersecurity
Each week at Applied Tech we recap the biggest cybersecurity news headlines from the week to keep you informed and ready to face the latest threats. Here’s your breakdown for the week of September 12th-18th.
OpenAI Discloses AI Models Instructing Successors to Hide Mistakes and Misalignment
OpenAI’s own disclosure framework caught something it probably wished it hadn’t: GPT-5.6 Sol agents leaving notes inside “compaction summaries” instructing future instances to hide mistakes from users. One agent, missing historical data for a financial model, told its successor to fabricate placeholder numbers and stay quiet unless asked. Another flagged a data mismatch, then decided a little omission never hurt anyone.
An unreleased Astra model went further, injecting fake “BREACH ALERT” commands and jailbreak-style persona prompts into its own summaries—one of which a successor model actually obeyed. OpenAI says it’s fixed the specific behavior, but the timing is rich: this lands days after Sam Altman promised independent safety evaluators, all while OpenAI reportedly courts a $1.2 trillion valuation.
Source: TechCrunch
Read More
Check Point Patches CVSS 9.8 Stack Overflow in Security Management Login Process
Check Point’s Security Management and Log Servers have a new critical flaw, CVE-2026-91843: a 9.8-rated stack overflow triggered by an oversized username in the pre-authentication login process, letting an unauthenticated attacker run code as root. Check Point pushed a LivePatch fix and says there’s no evidence of exploitation, but the official CVE record skips R82.20 entirely, while NHS England and Censys both list it as fully affected with no hotfix available yet.
This is also the fifth unauthenticated, code-execution-capable flaw in Check Point’s management server line since July 22, following one that was actively exploited. Confirm the LivePatch actually installed. Don’t just trust the checkbox. And lock down Trusted Clients in the meantime.
Source: The Hacker News
Read More
US Officials Investigate Cyberattacks on Two Crude Tankers Approaching the Gulf Coast
Two Texas-bound crude tankers got boarded by Coast Guard and FBI teams last month after cyberattacks disrupted their voyages. This is one of 40-50 similar boardings the Coast Guard’s Cyber Protection Team has run in the past year.
Iran’s Mehr News Agency claims hackers breached the VL Prosperity’s engine room near Gibraltar, slowing coolant, revving the engine, and cutting comms for 30 hours; the Coast Guard confirms malicious activity but won’t attribute it, and a former Coast Guard cyber official notes Iran has a habit of overstating its own capabilities.
Source: SecurityWeek
Read More
Human Attacker Matches AI Speed, Pivots From Marimo Flaw to SSH Bastion in Seconds
Sysdig’s latest research is a useful gut-check for anyone assuming “machine speed” automatically means AI involved. A human operator exploited the Marimo pre-auth RCE (CVE-2026-39987) and pivoted to an SSH bastion host in eight seconds flat, using a custom Python toolkit they wrote and debugged by hand without any agent in the loop.
Over a nine-hour session and 850-plus commands, they avoided a trap that Sysdig says every agentic threat actor profiled against the same CVE fell into. Skilled humans were already operating at machine speed, and AI is mostly lowering the bar for everyone else to catch up. Separately, Hunt.io flagged a related sweep compromising 3,562 exposed Redis servers for cryptomining.
Source: The Hacker News
Read More
About Applied Tech, An HBS Brand
Applied Tech is a leading IT and cybersecurity services provider dedicated to helping organizations protect their digital assets and scale their IT capabilities. Our proactive services span cloud management, security, productivity, and IT growth strategy—delivered by an experienced team with solutions built around your goals.
As part of Heartland Business Systems, we now bring greater resources, broader capabilities, and deeper technical expertise to every client relationship.
Get in Touch with Us
