A clearer view of today.

Researchers report a mass credential-exposure incident dubbed “FortiBleed,” with plaintext admin passwords appearing online for tens of thousands of internet-facing Fortinet devices. The scale (tens of thousands; claims extend to hundreds of thousands/half the exposed base in some reporting) turns a “credential leak” into an immediate, operational attack risk: attackers can authenticate directly and move from access to persistence quickly. The key takeaway for defenders is to treat exposed management credentials as an incident category of their own—rotate immediately, verify device integrity, and review for unauthorized configuration changes.

Source: Security Affairs

Microsoft confirmed work on a fix for a Windows Defender/Microsoft Malware Protection Engine elevation-of-privilege zero-day tracked as CVE-2026-50656. Reported exploitation involves a Defender exploit path where improper link resolution before file access can lead to SYSTEM-level outcomes, reportedly without requiring user interaction. Organizations running Defender should treat this as “patch now, hunt later,” prioritizing systems with the relevant Defender configuration and monitoring for suspicious privilege escalation patterns.

Source: SecurityWeek

New reporting describes DragonForce operators using Microsoft Teams relay servers to route command-and-control traffic, helping the malware blend in with legitimate enterprise communications. This technique can complicate network detection and reduce the effectiveness of traditional indicators that assume direct outbound C2 patterns. The actionable insight: expand detection beyond “known bad endpoints” to include anomalous application-layer behaviors and investigate Teams-related relay usage where it doesn’t match normal baselines.

Source: SecurityWeek

Researchers found an exposed dataset described as containing roughly 24 billion stolen records, including usernames and passwords collected from multiple sources. Even if not all entries are unique or active, the volume is large enough to materially increase the odds of successful credential stuffing and downstream takeover attempts. Defensive priority: enforce MFA, reduce password reuse blast radius through forced resets where appropriate, and consider account-level monitoring focused on login anomalies.

Source: Malwarebytes Blog

A report examined attacker sessions leveraging Claude Code and OpenAI Codex agents and found that most guardrails can be bypassed with surprisingly little skill. The implication is not just “AI increases attacker capability,” but that agentic tooling may lower the skill floor for multi-step intrusion workflows and accelerate iteration against targets. Defenders should assume faster weaponization cycles and invest in stronger identity controls, tighter execution environments for developer/ops tools, and better detection of agent-driven behaviors.

Source: Help Net Security

Coverage of OpenClaw (an open source, self-hosted AI assistant that can integrate with messaging platforms and take actions on local systems) emphasizes how autonomous agent capabilities create high-impact security risks when deployed without safeguards. The article notes concerns including credential/data exposure, indirect prompt injection, third-party skills supply-chain compromise, excessive permissions, and difficulty stopping an agent once it begins destructive actions. The key message for CISOs: focus on governance controls for agent identity and permissions, audit skills provenance, and require approval for actions with real-world effects.

Source: TechTarget