Hi there, this is your daily ☕️ Cyberpresso.
In today's Cyberpresso:
📮 French postal hack exposes 22M people
🤖 Amazon traced AI agents stealing LLM API keys
⛏️ Hackers hijack Bitcoin miners for rewards
🐧 Linux flaw crashes servers remotely
🚨 IRC server flaw lets hackers run commands
Plus: 💡 4 strategies & tactics, 🎁 8 more stories you might like, 🧰 6 tools, and 📚 5 papers.
French postal hack exposes 22M people LINK
- A September cyberattack on Maileva, a subsidiary of France's La Poste, leaked the names and postal addresses of 22 million people, including hospital patients and social housing tenants, in one of the country's largest breaches.
- Maileva prints and mails documents like payslips, invoices and payment reminders for over 70,000 client companies, so those exposed were not its customers but the people who received those letters and often never knew the firm existed.
- No bank details were taken, but a hacker reportedly read payment reminders from Foch Hospital subcontractor Kiwee.care showing invoice numbers, treatment dates and amounts owed; Maileva has told the CNIL regulator and filed a criminal complaint.
Amazon traced AI agents stealing LLM API keys LINK
- Amazon's threat intelligence team used decoy servers to catch attacker AI agents hunting internet-exposed AI apps like LiteLLM, One API, and Dify to steal the stored provider keys that powered their further searches.
- The agents, running with moderate confidence on the open-source pentest tool ARTEX and tied to China-linked groups, targeted hundreds of systems using default credentials, exposed keys, and known flaws in Langflow (CVE-2025-3248, a KEV-listed bug) and RAGFlow.
- Amazon could not verify the agents' reported results, but recommends treating AI gateways like production systems, changing default keys such as LiteLLM's master key, rotating exposed secrets, and watching for odd spending or usage.
Hackers hijack Bitcoin miners for rewards LINK
- Hackers broke into Luxor's Commander mining-management platform over the weekend and rerouted thousands of Bitcoin miners to the OCEAN pool, sending rewards to a single wallet while owners kept paying for electricity.
- The attack began around 8 p.m. Central on Saturday and allegedly redirected about 64,000 machines and 13 EH/s of computing power, roughly 1.5% of Bitcoin's global hashrate, with one flagged wallet collecting three payouts totaling 0.24 BTC.
- Luxor patched Commander within two hours, started restoring machines, and promised full compensation; CEO Nick Hansen blamed a "sophisticated AI-enabled vector" but has not revealed the exact entry point, pending a full post-mortem.
Linux flaw crashes servers remotely LINK
- A Linux kernel bug named NoisyTCP, hidden for 11 years, lets an attacker with no login crash any server remotely by sending crafted network packets that force the system to panic.
- The flaw, tracked as CVE-2026-99288, sits in the kernel's TCP code and puts default Nginx, Apache, and SSH servers at risk, with a crash in a container taking down the entire host.
- Researcher Hyunwoo Kim patched it and released an advisory on GitHub; admins should install vendor kernel updates and reboot, or temporarily turn off SYN cookies or cut the interface to one receive queue.
IRC server flaw lets hackers run commands LINK
- CISA has flagged a serious bug in UnrealIRCd, an open-source Internet Relay Chat server, warning that hackers are already abusing it to run commands with server-level powers on affected networks.
- The flaw, CVE-2026-114713, lets any user on a network with two or more linked servers inject server-to-server commands, because UnrealIRCd fails to escape the IRCv3 "label" message tag when echoing replies.
- Maintainer Bram Matthys released the patched version on November 7, calling it the worst bug in 15 years; exploitation began a day later, and federal agencies had until today to patch or deploy a spamfilter workaround.
Strategies & Tactics
> A Valid Signature Isn’t Enough: Attributing and Validating Leaked JWTs: Validating a leaked token against its own service's API, not just checking its cryptographic signature, reveals that over 334,000 of 2.4 million tokens still work.
> The Green Agent POV: Formulating the Fix for an Internet-Facing Container with Sensitive Data Access: Wiz's Green Agent investigates a critical cloud risk in minutes, recommending immediate fixes that neutralize threats without taking down production services.
> Adapting to AI penetration testing orchestrators: The Modern Security Model: Deploy a layered defense combining firewalls, bot detection and identity-gated access because AI-driven hacking tools now attack at machine speed and scale.
> From Report to Tested Detection in Hours: The ARTEX Signature Pack: Detect AI-driven attacks by their behavior patterns rather than easily-changed indicators, so defenses stay valid even when attackers swap models, prompts or servers.
Other news you might like
- A USB worm’s second life: Retadup spreading CoffeeLoader LINK
- Hackers Use SANSAR Linux Malware to Hide Crypto Mining by Faking 0% GPU Usage LINK
- Tor releases emergency update as DDoS attack floods relays LINK
- Hackers Probe Newly Disclosed Flaws in Copeland Refrigeration Controllers LINK
- New Android Malware Hides Hacker Activity While Stealing Banking Data LINK
- Apple Fixes iOS Flaw That Let Sandboxed Apps Secretly Record Microphone Audio Without Permission LINK
- BlackNevas Ransomware Wipes VMware ESXi Infrastructure and Backups Without Exploiting Any Flaws LINK
- Cursor Security Breach Exposes API Credentials and Source Code Following Fireworks AI Supply Chain Attack LINK
Trending tools
VibeDefend: embeds security and business rules into AI coding agents like Cursor and Copilot, catching vulnerabilities and malicious packages before code reaches your PR. LINK
Ghostifier: scans inbox headers to find companies holding your data, then unsubscribes and sends tracked opt-out and deletion requests without reading emails. LINK
Upsolve Row-Level Security: enforces data access permissions server-side, letting AI agents safely serve multiple tenants without exposing or bypassing row-level rules. LINK
Off the Record: emits inaudible interference that blocks AI notetakers from transcribing your meetings, including hidden ones running silently on other people's devices. LINK
Koreshield: screens customer messages, retrieved documents, and tool calls before your AI agent acts, catching data leaks, hidden instructions, and unsafe actions while logging decisions. LINK
AuthMonster: handles two-factor authentication with biometric locking, simple migration from Google Authenticator and LastPass, and a donation-based model without paywalls. LINK
Trending papers & reports
Camera fingerprint forgery can be faked so convincingly that it fools standard photo-origin tests while staying visually flawless, but a free detection toolkit caught every forgery, showing courtroom-style image evidence needs backup checks. LINK
Chip speculation leaks get a targeted fix that adds protection only where secrets could actually spill during a processor's fast-guessing shortcuts, cutting the usual slowdown while mathematically proving the guarded code stays secure. LINK
Cache harm ranking pinpoints which shopper sessions quietly knock reused items out of a store's web cache, finding the worst 1% cause ~49% of the damage and half carry no bot label. LINK
Image watermarking upgrades let companies swap or combine different invisible-watermark systems by attaching a small add-on instead of retraining from scratch, using as few as 1 to 2 million adjustable settings per piece. LINK
Rogue AI agents can poison a team of AI workers by feeding bad answers to their assigned tasks, so this new defense picks the response most agreeing with others, blocking the attacks without hurting performance. LINK
See you tomorrow for a new dose of ☕️ Cyberpresso!
