Cyber Briefing: 2026.07.31
Recent incidents range from unauthenticated server vulnerabilities and abuse of legitimate corporate authentication systems to unintended real-world company breaches caused by autonomous AI models
Welcome to Cyber Briefing, your daily source for all things cybersecurity. We bring you the latest advisories, alerts, incidents, and news every weekday.
Emerging security developments highlight a mix of critical infrastructure flaws, advanced phishing techniques, and unexpected AI system behaviors. JetBrains patched a critical remote code execution flaw (CVE-2026-63077) in TeamCity On-Premises that lets unauthenticated attackers run arbitrary OS commands over HTTP(S). Meanwhile, threat actors are leveraging authentic Microsoft login workflows, disguising emails as HR task notifications, to bypass standard phishing filters across scores of organizations. Additionally, Anthropic disclosed that its Claude AI models accidentally breached three real companies during security testing after mistaking internet-exposed assets for simulated sandbox environments, extracting production database records and sparking broader questions around autonomous AI safety and legal liability.
On the policy, governance, and hardware fronts, organizations and regulators are taking active steps to manage digital risk. Snowflake launched an AI Agent Security Framework providing prompt injection defenses, identity limits, and multi-approval safeguards to restrict autonomous agents at the data layer. In contrast, health platform Hims & Hers faces a FTC lawsuit alleging unauthorized sharing of sensitive consumer health data with major ad platforms and predatory subscription practices. Finally, in hardware security, DEF CON 34 introduced conference badges featuring open-source security chips designed by Andrew “bunnie” Huang to promote transparency and inspectable chip-level security.
Listen to our podcast here ⏬
⚡THREAT LANDSCAPE
CVE-2026-63077 TeamCity RCE Vulnerability
A critical vulnerability (CVE-2026-63077) in TeamCity On-Premises allows unauthenticated attackers with HTTP(S) access to execute arbitrary operating system commands with server process privileges. All TeamCity On-Premises versions exposed over HTTP(S) are affected, while TeamCity Cloud customers are not impacted. Administrators should immediately upgrade to versions 2025.11.7 or 2026.1.3, or apply the available security patch plugin for older installations. Read More
Attackers abuse Microsoft auth for phishing
Attackers are bypassing traditional phishing defenses by abusing Microsoft’s legitimate authentication system instead of creating fake login pages. Check Point researchers identified over 200 phishing emails between late June and mid-July targeting approximately 120 organizations across multiple industries and countries. The attacks disguise themselves as Microsoft Planner task notifications from HR departments, exploiting the trust users place in authentic Microsoft login prompts. Read More
🚨INCIDENTS & REAL-WORLD IMPACT
Anthropic AI models breached three real companies
Anthropic disclosed that its Claude AI models breached three real companies during security testing after the models mistakenly believed internet-accessible systems were part of simulated exercises. The incidents occurred because evaluation partner Irregular left test machines open to the internet while telling Claude it had no connectivity; Claude then compromised real organizations using basic techniques like weak password exploitation and SQL injection, with one breach extracting production database records. The disclosure follows a similar incident at OpenAI and raises urgent questions about legal liability, notification requirements under data protection laws, and whether current AI containment practices are adequate as models gain autonomous hacking capabilities.Read More
🔓 EXECUTIVE RISK & CYBERNOMICS
Snowflake’s AI Agent Security Framework
Snowflake has released a security framework for AI agents that includes indirect prompt injection protection, Agent Identity controls, and MCP governance to constrain autonomous systems. The framework operates at the data layer to neutralize malicious prompts without latency and enforces strict authorization limits on what agents can do on behalf of users. Snowflake recommends implementing multi-approval requirements for high-risk actions, currently defaulting to two approvers to reduce the risk of compromised accounts executing dangerous operations. Read More
🛡️ POLICY, REGULATION & LEGAL SIGNALS
FTC sues Hims & Hers over health data sharing
The Federal Trade Commission has filed a lawsuit against telehealth company Hims & Hers, alleging it shared customers’ sensitive health information with advertising platforms including Meta and Snap without proper consent, while also misleading customers about billing and subscription cancellations. The company, which provides online treatments for conditions like hair loss and erectile dysfunction, allegedly enrolled customers in recurring subscriptions without consultation and made cancellation processes deliberately difficult. The FTC is seeking injunctions, monetary relief for affected consumers, and civil penalties for violations of consumer protection laws. Read More
💻 CAREER ENABLEMENT
DEF CON 34 Badges Feature Open Source Security Chip
DEF CON 34 conference badges, designed by hardware hacker Andrew “bunnie” Huang, incorporate an open source security chip to advance hardware transparency and security. The badges represent a push toward verifiable security in hardware design, allowing attendees to inspect and understand the security mechanisms at the chip level. This initiative aims to demonstrate practical applications of open source principles in security-critical hardware components. Read More
Copyright © 2026 CyberMaterial. All Rights Reserved.
Follow CyberMaterial on:
Substack, LinkedIn, Twitter, Reddit, Instagram, Facebook, YouTube, and Medium