☕️ Millions of AWS resources briefly exposed

Hi there, this is your daily ☕️ Cyberpresso.

In today's Cyberpresso:

☁️ Millions of AWS resources briefly exposed

🔓 Wallet bug drains $38M in Bitcoin

🏥 Health billing firm leaks 350K records

🔎 Google's AI fixed 1072 Chrome bugs

🚰 Hackers hit water systems in 7 states

Plus: 💡 5 strategies & tactics, 🎁 7 other news you might like, 🧰 6 tools, and 📚 5 papers.

☁️ Millions of AWS resources briefly exposed LINK

Researchers at Aryon disclosed "Cloud ShutterGap," a blind spot where millions of AWS resources, including RDS and DocumentDB snapshots, Amazon Machine Images, and Systems Manager documents, are briefly exposed to the public before being deleted, often within minutes.

Because CSPM and CNAPP tools rely on periodic scans, exposures that appear and vanish between scan cycles go undetected: 20% of public RDS snapshots lasted under two minutes and 99% of deleted snapshots were gone within 30 minutes.

Attackers who continuously enumerate public AWS resources can copy an exposed snapshot into an account they control within seconds, retaining data, account IDs, emails, private keys, potential payment card data, so Aryon recommends prevention via Service Control Policies, encryption, and account-level public-sharing blocks.

🔓 Wallet bug drains $38M in Bitcoin LINK

Coinkite is urging Coldcard Mk3 users to move their Bitcoin after warning that wallet seed phrases generated on the device may have been created with insufficient randomness, as investigators track a sweep of 594.48 BTC worth roughly $38.3 million.

The flaw affects seeds made on Mk3 firmware version 4.0.1 through 5.0.3, though no public evidence yet confirms it caused the sweep of 1,324 unspent outputs across 500 transactions in a three-block window; the Mk4, Q and Mk5 are not affected.

Researchers suspect a low-entropy random-number generator let an attacker brute-force single-signature wallets, possibly limited to BIP-84 paths, so Coinkite says seeds using a BIP-39 passphrase face minimal risk and advises generating a new seed on an unaffected device.

🏥 Health billing firm leaks 350K records LINK

Healthcare IT company CareCloud is notifying at least 350,000 people that hackers stole their personal, financial, and medical information in a data breach involving an electronic health record environment in its CareCloud Health division.

Attackers accessed one of CareCloud's AWS environments between March 10 and March 16, 2026, and likely exfiltrated data before the environment was disrupted on March 16, with the theft confirmed by the company's June 24 investigation.

The stolen data includes names, addresses, Social Security numbers, dates of birth, driver's license numbers, government ID numbers, financial account numbers, credit/debit card numbers, and medical and health insurance information, and CareCloud is offering 24 months of identity theft protection.

🔎 Google's AI fixed 1072 Chrome bugs LINK

Google's Chrome security team shipped fixes for 1,072 security bugs across two major releases in June, more patches than the team delivered in the previous 23 big releases combined, driven largely by AI tools finding vulnerabilities.

Chrome has used machine learning to find vulnerabilities and automate fuzz testing since 2012, but the team now trains its AI model on every past CVE and every line of code in Chromium's history.

The team is piloting security fixes twice a week to handle the surge, though Tabriz and Turner expect a new equilibrium as AI finds the bulk of bugs and structural fixes like Rust rewrites remove whole bug categories.

🚰 Hackers hit water systems in 7 states LINK

The FBI and EPA issued a joint warning after hackers hit water and wastewater utilities in at least seven US states, disrupting operations since July 27, 2026 by breaking into internet-facing programmable logic controllers.

The attackers gained remote access to exposed Rockwell Automation Allen-Bradley MicroLogix 1100 and 1400 PLCs and changed device settings, including IP addresses and passwords, causing operators to lose monitoring and control over affected systems.

The disruptions included pressure loss and flooding, with reduced pressure potentially letting untreated groundwater enter pipelines; the agencies advise disconnecting PLCs from direct internet access and using secure gateways, firewalls, and strong unique passwords.

💡 Strategies & Tactics

> Anthropic's Claude breached 3 orgs, uploaded PyPI malware during tests: A misconfigured test that falsely told Claude it was offline let the model breach three real organizations, showing that sandbox failures, not model malice, are the danger.

> Critical SolarWinds Flaw Lets Attackers Bypass Web Help Desk SAML Login: A critical SolarWinds flaw lets attackers skip single sign-on and reach Web Help Desk data, so administrators should upgrade to version 2026.2.1 immediately.

> SilverFox Targets Japanese Manufacturer With Advanced ValleyRAT Campaign: SilverFox hijacks trusted PDF apps to run malware that can kill security tools and self-restart, so blocking one piece won't stop the attack.

> ‘DangleGeddon’: AI Could Weaponize Forgotten DNS Records at Global Scale: AI lets attackers rapidly find and hijack forgotten DNS records pointing to deleted cloud resources, turning a niche exploit into mass disruption that nation-states could weaponize.

> Ransomware Killers Overwrite Security Process Memory Without Terminating Applications: New ransomware quietly edits the memory of security tools instead of killing them, so dashboards show protection while detection is silently disabled during encryption.

Other news you might like

• Astaroth Banking Trojan Adds WhatsApp Web Spambot to Spread Malware Across BrazilLINK

• OctLurk and SilkLurk Backdoors Target Central Asian Governments in Cyberespionage CampaignLINK

• New GenieLocker Ransomware Encrypts Windows, Linux and VMware ESXi SystemsLINK

• Chinese-Speaking Hacker Uses DeepSeek Agent to Launch Autonomous CyberattacksLINK

• BlackTech APT Uses New BlueShell Linux Backdoor in Attacks on Japanese OrganizationsLINK

• The Gentlemen Ransomware Kills Nearly 180 Security Processes Before Encrypting Your FilesLINK

🧰 Trending tools

Perfai Security: automatically scans and fixes access control vulnerabilities in apps built with AI coding tools like Replit, Lovable, and Cursor, no security expertise required.LINK

Constellation Gate AI: routes AI agent traffic through a security gateway that blocks prompt injection, scans for secrets, and reduces token costs 20-40% via compression.LINK

Gander: an Android file viewer that opens PDF, Word, Excel, PowerPoint, images, video, audio, and code files without requesting any permissions.LINK

Claudoscope: a free, open-source macOS menu bar app that browses Claude Code session history, tracks token costs, and scans for leaked credentials locally.LINK

darknet-mcp-server: a security research tool offering 66 functions for dark web monitoring, including breach detection, ransomware tracking, Tor access, and malware analysis.LINK

recon-skills: a collection of 162 tested offensive security techniques for reconnaissance and penetration testing, covering subdomain takeovers, secret hunting, and system enumeration methods.LINK

📚 Trending papers & reports

Poisoned training data can be cleaned using an outside image-recognition model as a judge, cutting successful hidden-backdoor attacks to under 1% while barely touching accuracy, a drop of just 0.3%.LINK

Security checklists for AI features fail to catch major risks, like software supply-chain gaps and human-centered threats, when tested on a real small-business system, showing why standard threat modeling needs a GenAI-specific overhaul.LINK

Content moderation filters for images can be tricked by basic edits like color inversion or grayscale, letting harmful content slip past all three major commercial safety APIs tested.LINK

Permission checks for AI agents mathematically verify each action against security rules before it runs, blocking unsafe steps with a precise explanation and hitting a ~90% compliance rate across test scenarios.LINK

Cloud IoT access rules can be automatically checked for hidden data leaks between devices, using a tool that maps allowed communications and flags unsafe information flow before attackers exploit it.LINK

See you tomorrow for a new dose of ☕️ Cyberpresso!