☕️ Hackers exploit critical Oracle flaw

Hi there, this is your daily ☕️ Cyberpresso.

In today's Cyberpresso:

🔓 Hackers exploit critical Oracle flaw

💰 "OkoBot" malware steals crypto wallet seed phrases

📌 OpenAI Model Builds Chrome Exploit

🪟 New Windows exploit beats latest patches

Plus: 💡 5 strategies & tactics, 🎁 8 other news you might like, 🧰 6 tools, and 📚 5 papers.

🔓 Hackers exploit critical Oracle flaw LINK

Attackers are actively exploiting a critical flaw in Oracle Payments, a component of Oracle E-Business Suite, to take full control of payment-processing systems, prompting CISA to warn organizations to remediate immediately.

The bug, tracked as CVE-2026-46817, is an improper privilege management issue that lets an unauthenticated attacker with HTTP network access completely take over Oracle Payments, with no valid login required, especially where the service is exposed to the internet.

CISA confirmed real-world exploitation, giving federal agencies until July 18, 2026, to fix it, and advises applying Oracle's fixes, reviewing logs for suspicious HTTP activity and new accounts, or discontinuing use if no mitigation is available.

💰 "OkoBot" malware steals crypto wallet seed phrases LINK

Kaspersky uncovered an active malware campaign dubbed "OkoBot" that steals cryptocurrency wallet files, browser cookies, and login credentials from infected Windows machines through an SSH tunnel to attacker-controlled servers.

The still-active campaign starts with a malicious PowerShell downloader called TookPS, spread via ClickFix attacks and GitHub repositories posing as legitimate software, such as a fake SQL Server Management Studio package that was actually the audio editor Audacity with a hidden implant.

Once in, an automated SSH bot harvests wallet files and credentials, disables Windows Defender notifications, opens RDP access with a patched termsrv.dll, and deploys over 20 modules including a launcher that bypasses UAC via msconfig.exe and the Rilide browser stealer that targets cryptocurrency theft.

📌 OpenAI Model Builds Chrome Exploit LINK

OpenAI's GPT-5.6 Sol Ultra model independently built a complete, working exploit chain for Google Chrome, starting from nothing but publicly available V8 security-fix commits, in a test run by researchers at Hacktron.

Working against the V8 source at version 14.9.207.35 (matching Chrome 149.0.7827.201) in a sandboxed test build, the model chained a type-confusion bug and a use-after-free to escape the sandbox and run code, popping a calculator as proof.

The full run used 2.1 billion tokens across 14,062 requests for about $1,597, and Hacktron warns that letting well-resourced attackers throw compute at weaponizing patches this way could shrink the patch gap, making known unpatched (N-day) bugs a bigger risk.

🪟 New Windows exploit beats latest patches LINK

A newly released proof-of-concept exploit called LegacyHive targets a Windows elevation-of-privilege flaw in the Windows User Profile Service, and it still works on systems that have the July 2026 security updates installed.

The flaw lets a standard user load another user's registry hive under their own registry classes root, and by mounting a privileged account's hive an attacker can find escalation paths from the local system configuration and accessible registry data.

The public PoC by Nightmare-Eclipse is restricted, needing a second standard user's credentials and a third account username; there is no CVE, Microsoft advisory, or patch yet, so administrators should restrict local access and monitor for unusual loading of NTUSER.DAT and UsrClass.dat.

💡 Strategies & Tactics

> The Risk of Exposed Cloud Functions and How to Harden: Isolate public serverless functions and enforce least-privilege access so that a single code vulnerability cannot escalate into a full cloud takeover.

> Investigating Persistence Mechanisms in AWS: Explains how to detect and shut down attacker footholds in AWS by hunting for suspicious IAM users, backdoored role policies, and malicious Lambda functions in cloud logs.

> There and Back Again: An Operators Guide on NTLM Relaying Egress: Explains how attackers escalate privileges when firewalls block internal relay attacks by coercing NTLM login traffic out to a cloud server and proxying it back into the target network.

> The Red Agent POV: The One Boolean That Broke a B2B Platform’s Credit System: An autonomous AI hacking agent unlocked a platform's entire paid contact database for free by flipping one client-side flag the server never verified.

> Claude for Chrome flaw could let rogue extensions access your Gmail: A flaw lets malicious browser extensions impersonate you and command Claude's Chrome assistant to read Gmail or send emails without your knowledge, and Anthropic's fix remains incomplete.

Other news you might like

• Zoom warns of critical account takeover vulnerabilityLINK

• New Spirals ransomware encrypts victim network in under 24 hoursLINK

• Russian hackers trojanize WebEx, Zoom apps to push Starland malwareLINK

• OpenAI is now using AI to attack its own AI, and it's working better than humans ever didLINK

• Hackers Pair Stolen Wallet Databases With Keychain Passwords for Offline Crypto TheftLINK

• Windows Bind Link Attacks Can Hide Malware From EDR ToolsLINK

• CISA warns that multiple vulnerabilities in SharePoint are under exploitationLINK

🧰 Trending tools

Constellation Gate AI: routes AI agent traffic through a gateway that blocks prompt injection, scans for leaked secrets, and logs verifiable audit trails while cutting token costs 20-40%.LINK

TailMux: lets you connect to multiple Tailscale tailnets simultaneously on macOS and Linux by running isolated embedded nodes per profile, routing by hostname without account switching or VMs.LINK

Claudoscope: a free macOS menu bar app that browses Claude Code session history, tracks token costs, scans for leaked credentials, and lints CLAUDE.md configs locally.LINK

Netfox: a native macOS app that scans your network via Bonjour, ARP, SSDP and NetBIOS to track every connected device, flag risks, and log history locally.LINK

DeepFrame: a security studio running authorized deep penetration tests for fast-moving web apps, delivering clear findings and free retests.LINK

Origin: a confidential agentic stack for regulated industries, offering a private LLM gateway, AI IDE, agents, attestation, and sandboxes for secure workflows.LINK

📚 Trending papers & reports

Turning mockups into code got 46.7% to 69.4% faster when AI tools were combined with a company's design rules, an enterprise experiment found.LINK

Code review teams mixing humans, AI reviewers, and AI agents get faster decisions but not better quality, a study of 1.02 million pull requests across 207 GitHub projects finds.LINK

Multi-level modelling tools struggle when AI chatbots try to build them, getting the stated rules right but missing unstated structure, with correctness scores ranging only from 52% to 79%.LINK

Chat-to-code gets more accurate with a new filtering method, improving precision by up to 20 points on average across 541 test tasks compared to using AI translation alone.LINK

Software updates for long-running programs can silently break tasks already in progress, so this method scores upgrade risk from existing logs alone, sorting runs into safe, review, or pause groups without extra test runs.LINK

See you tomorrow for a new dose of ☕️ Cyberpresso!