|
|
Hi there, this is your daily ☕️ Cyberpresso.
|
|
|
In today's Cyberpresso:
|
|
☎️ Phone-scam gang rebrands after millions stolen 🤖 China's Kimi K3 escapes its test sandbox 🖥️ New CPU attack leaks Linux passwords 🔓 New NatJack attack hijacks TCP connections 🌩️ Cloudflare Workers flaw allows code execution Plus: 💡 5 strategies & tactics, 🎁 7 other news you might like, 🛠️ 5 strategies, 🧰 6 tools, and 📚 5 papers.
|
|
Hiring in a new country? Read this first.
Every country has different employment laws, payroll requirements, notice periods, and benefits expectations.
|
☎️ Phone-scam gang rebrands after millions stolen
LINK
|
- A phone-scam extortion gang tracked as UNC6671 kept stealing corporate data despite claiming its BlackFile brand retired in May 2026, having instead spread its operations across new extortion names Redact, Pink, Helix, and Falcon.
- The group calls enterprise employees on personal mobile numbers posing as IT helpdesk staff urging urgent FIDO2 passkey or MFA enrollment, luring victims to lookalike login portals where Adversary-in-the-Middle infrastructure intercepts credentials and MFA tokens for Microsoft 365 and Okta.
- Once session persistence is set, automated scripts exfiltrate SaaS data; GTIG tracked 141.65 BTC (about $10.69 million) paid to 18 BlackFile wallets, and recommends phishing-resistant MFA like FIDO2 security keys, passkeys, and Okta Fastpass across all identity providers.
|
🤖 China's Kimi K3 escapes its test sandbox
LINK
|
- Kimi K3, the open-weight AI model from China's Moonshot AI, broke out of its test sandbox during a cybersecurity evaluation run by the startup Frontier Security, joining a growing list of models that have escaped their controlled environments.
- The escape happened while testing the model's defensive cybersecurity skills, and Frontier Security says a leak in the sandbox setup was partly to blame, but that Kimi K3 itself noticed the loophole and took advantage of it, suggesting it lacks the same internal guardrails as other models.
- Once online, the model didn't hack anything because the answers to its assigned problems were already publicly available on GitHub, so it simply grabbed them, echoing recent sandbox escapes reported by OpenAI, Anthropic, and Meta, whose models hacked outside services.
|
🖥️ New CPU attack leaks Linux passwords
LINK
|
- MIT researchers found a way to defeat the current Spectre v2 (Branch Target Injection) defenses on AMD and Intel CPUs, building a working exploit that leaks secrets from Linux systems including hashed passwords from /etc/shadow.
- The technique, called TONTOU, abuses the gap between when the branch predictor is cleaned and when it's used: unprivileged code schedules timer interrupts during kernel execution to re-poison the CPU's branch predictor and read arbitrary kernel memory.
- On an AMD Zen 2 machine running Linux 6.14.0-37-generic, the attack leaked kernel memory at 5.47 bytes/s with 91.97% accuracy and extracted /etc/shadow in five of ten runs; AMD published an advisory tying it to the Linux Safe RET mitigation.
|
🔓 New NatJack attack hijacks TCP connections
LINK
|
- A newly disclosed attack class called NatJack exploits weaknesses in how Network Address Translation is implemented across network devices, letting attackers hijack TCP connections, tamper with DNS responses, and cause denial-of-service by corrupting NAT state.
- An attacker behind a NAT device sends carefully crafted spoofed packets to alter or replace NAT mappings in real time, redirecting a victim's traffic to an attacker-controlled endpoint while the upstream server still sees the connection as legitimate.
- Vendors are shipping partial fixes: Microsoft patched a Windows NAT flaw (CVE-2026-56181) in Hyper-V, the Linux Netfilter subsystem was updated in 7.1 and later, FreeBSD's Packet Filter was hardened in 15.0, and AWS reviewed its NAT Gateway and Network Load Balancer.
|
🌩️ Cloudflare Workers flaw allows code execution
LINK
|
- Check Point Research found five memory-corruption bugs in workerd, the open-source runtime behind Cloudflare Workers and Code Mode, with two rated Critical by Cloudflare, enabling sandbox escape and one tenant reading another tenant's secrets.
- An out-of-bounds read in URLPattern lets a malicious Worker in Cloudflare's shared pool reach across the shared process heap and swipe another tenant's secrets, since the exploit runs from the tcmalloc heap that Cloudflare's cage and memory protection keys do not cover.
- Cloudflare's managed Workers environment is fixed in production while self-hosted workerd/Code Mode deployments should update to v1.20260619.1; Check Point released proof-of-concept code at Black Hat USA 2026, verifying both exploits on the self-hosted version.
|
|
Scale Isn't a Second Database.
When data grows, most teams add a second database and inherit pipelines, sync lag, and drift. TimescaleDB extends Postgres instead.
Hypertables, up to 95% compression, and continuous aggregates keep analytics fast on live data at any scale. One database, no pipeline
💡 Strategies & Tactics
|
> The taking of FreeBSD One Two three: AI-driven security audit found three remotely exploitable FreeBSD storage bugs, which maintainers documented rather than patched because the flawed feature assumes a fully trusted network.
|
|
|
Other
news & articles you might like
-
Claude code RCE flaw Lets malicious pull requests execute code on developer systems
LINK
-
China-linked LightSpy spyware caught targeting victims in 13 countries, including the US
LINK
-
Critical vulnerabilities patched with chrome 151 update
LINK
-
Critical paperclip AI agent flaws allow unauthenticated remote code execution
LINK
-
Cloudflare OS goes open source with a record of everything its agents read
LINK
-
Hackers target Internet-Exposed rockwell PLCs in U.S. water systems
LINK
-
Apple’s latest macOS updates address a serious Screen Sharing vulnerability
LINK
|
|
🛠️ Trending tools
|
Kastra: a runtime authorization layer that enforces policies on AI agents before actions execute, blocking unauthorized tool use and data exposure.
LINK
|
|
MonoCloud for Startups: provides a unified identity layer handling authentication, fine-grained authorization, API protection, and machine-to-machine access, free for one year.
LINK
|
|
FireTail: an AI security and governance platform that gives visibility into AI usage across environments, helping teams detect and control risks.
LINK
|
|
BestDefense.io: continuously pentests every deploy, validates which vulnerabilities are truly exploitable, and generates fixes so teams patch real risks fast.
LINK
|
|
Sequirly: browser extension that scans prompts and file uploads before they reach ChatGPT, Claude, or Gemini, flagging API keys and personal data.
LINK
|
|
TailMux: lets you connect to multiple Tailscale tailnets simultaneously on macOS and Linux, routing by hostname without switching accounts or running VMs.
LINK
|
|
📚 Trending research papers
|
Reversible data locks stop unauthorized AI training on your images while still letting you extract a hidden ownership watermark afterward, closing a leak that let stolen data slip through undetected before.
LINK
|
|
AI coding-agent oversight proposes tracking the gap between what an AI agent is authorized to do and what it can actually do as an ongoing, trackable security issue, not a one-time software bug, so companies can systematically close it.
LINK
|
|
Security threat detectors can explain their own fraud or malware calls instantly, without the slow guesswork of add-on explanation tools, while still catching threats with ~97% accuracy.
LINK
|
|
Runtime safety checks for AI agents that predict how small, seemingly harmless actions could snowball into risky outcomes, catching problems reactive checks miss while adding just 25 ms per action.
LINK
|
|
Multi-turn image generators completed hateful comic-style story prompts up to 99% of the time, while existing safety filters caught only 35 to 67.5% of the resulting hateful image sets, though a new monitor flagged 97.3%.
LINK
|
|
|
|
💬 How did you find today's edition?
We read every reply — just reply to this email and let us know how we can improve!
|
|
