🔑 768 AWS root keys still live, 💳 Apollo Global breach hits SSNs, 🔓 Spring Security CVSS 10 LDAP flaw, 🤖 DeepSeek AI automates exploits, 🕵️ Grok zero-click prompt injection, 🦠 LockBit claims U.S. Bank

Truffle Security identified 768 AWS access keys that remain active and grant full administrative control over corporate cloud accounts. The credentials include 526 root access keys and 242 IAM user keys with AdministratorAccess permissions, re-verified on August 10 with 88% still authenticating. Sourced from public Git histories, Hugging Face datasets, Docker images, package registries, and CI/CD logs, the keys span a four-year exposure window suggesting they were never rotated. Full admin access enables data theft, crypto mining, ransomware deployment, and cloud billing fraud at scale.

Apollo Global Management disclosed that attackers accessed its cloud platforms between July 6 and July 10 through a social-engineering attack, exposing names, birth dates, mailing addresses, and Social Security numbers. Apollo manages over $700 billion in assets, meaning the breach touches a broad population of high-net-worth individuals and institutional investors. The combination of SSNs with full PII enables fraudulent account applications, tax fraud, and synthetic identity creation.

Spring Security CVE-2026-59270 is a CVSS 10.0 flaw in the embedded UnboundID LDAP server that unconditionally registers a well-known administrative bind DN and binds the listener to all interfaces. The UnboundIdContainer component, designed for development, can be spun up in staging or production-adjacent environments by Spring Boot auto-configuration, where any network-reachable attacker can authenticate with the default credential and gain full directory read/write access. The flaw is part of a batch of 91 Spring security advisories disclosed August 20 across Spring Framework, Spring Security, and Spring Cloud, affecting an estimated 209,569 software components.

A Chinese-speaking threat actor built an autonomous attack pipeline using DeepSeek AI through the Hermes Agent framework, automating reconnaissance, vulnerability research, exploit acquisition, and attack attempts against internet-facing infrastructure. Unit 42 identified the actor chaining DeepSeek's reasoning with terminal access to scan targets, find CVEs, pull public PoC exploits, and launch attacks in a single workflow. The demonstration confirms that LLMs can compress the timeline from target discovery to exploitation, requiring no custom malware or novel tooling.

Adversa AI disclosed a zero-click prompt injection technique called Cryptographic Context Injection targeting xAI's Grok web chat. When a user asks Grok to summarize a webpage, hidden instructions on an attacker-controlled page are decrypted and treated as trusted internal context, exfiltrating the user's name, location, subscription tier, and active conversation history. Tested against Grok 4.5 Fast, the attack completed without a confirmation dialog or visible warning, demonstrating that agentic AI browsing capabilities create a practical data-exfiltration vector.

[UPDATE] TrueConf remains actively exploited by Head Mare, with CISA confirming KEV catalog additions for CVE-2026-72529 and CVE-2026-72530 on August 21. The supply-chain attack replacing legitimate client installers with PhantomCore backdoor payloads continues to affect versions 5.3.x before 5.3.9, 5.4.x before 5.4.9, and 5.5.x before 5.5.5.

[UPDATE] Microsoft patched Entra ID CVE-2026-69836 (CVSS 10.0), a deserialization RCE in the cloud identity provider, with a server-side fix requiring no customer action. Organizations should verify their Entra ID tenants are patched and review authentication logs for anomalous activity since August 20.

[UPDATE] LockBit 5 claimed a cyberattack against U.S. Bank and threatened to publish stolen data by September 3. U.S. Bank said it found no indication of internal system compromise but is investigating.

A critical N-able Passportal vulnerability (CVE-2026-15580, CVSS 9.4) in the browser extension allowed any malicious website to extract long-lived authentication tokens through improper postMessage validation. Passportal is an MSP-focused platform, meaning one compromised admin vault cascades privileged credentials and customer infrastructure access across an entire client base. N-able patched the issue in version 3.49.6 within 24 hours.

The phishing toolkit now uses passkeys to maintain access after password resets, representing a shift in credential-theft techniques that bypasses traditional incident response workflows.

Watchlist: rotate any leaked AWS credentials immediately; patch Spring Security against the 91-advisory batch, prioritizing CVSS 9.x+ network-reachable issues; verify Entra ID tenants are patched for CVE-2026-69836; update TrueConf Server to 5.5.5 or later; ensure N-able Passportal extension is at version 3.49.6 or later; monitor LockBit's September 3 deadline for U.S. Bank data exposure; and evaluate passkey enrollment policies to prevent post-reset persistence by credential-theft toolkits.