Intelligence · Updated daily
AI-analysed threats, vulnerabilities and campaigns. Not just what happened — what it means, who's affected, and what to do about it.
Attackers are exploiting autonomous AI agents by injecting malicious content into trusted data sources, using techniques ranging from hidden code injection to cognitive state manipulation. This represents a new attack surface where data integrity failures can compromise AI decision-making at scale.
CISA has confirmed active exploitation of CVE-2025-67038, a code injection vulnerability in Lantronix EDS5000 Series devices with a CVSS score of 9.8 that enables remote code execution. Federal agencies are required to patch by June 26, 2026.
Microsoft and law enforcement coordinated action against three major cybercrime-as-a-service operations (Stealc, Amadey, Socgholish), taking down over 300 servers. The operation targets the malware supply chain rather than individual infections, representing a shift toward infrastructure disruption.
Google is introducing enhanced privacy controls allowing users greater granularity over activity history and personalisation settings across Search and Play services. This reflects incremental progress in user data governance rather than a response to a specific incident.
A chain of two authentication design flaws allows unauthenticated attackers to read the admin configuration file and reuse the SHA-1 password hash directly as an API signing key, bypassing authentication entirely when normal user password is left empty (default configuration).
Two Scattered Spider members pleaded guilty on day one of their UK trial for the August 2024 ransomware attack on Transport for London. The rapid guilty pleas suggest strong prosecution evidence and mark a significant enforcement win against a prolific cybercrime group.
Anthropic's Mythos model identified vulnerabilities in classified US government systems during testing, though exploitation capability within the testing window remains unclear. This highlights potential risks when deploying advanced AI models against sensitive infrastructure.
A Russian-speaking initial access broker has been conducting a sustained credential-harvesting campaign against FortiGate firewalls since February 2026, compromising over 430,000 devices globally and harvesting approximately 110 million credentials. This represents a significant threat to enterprise network security infrastructure and likely serves as a precursor to deeper compromise or sale of access.
The Five Eyes intelligence alliance has issued a joint alert warning that artificial intelligence presents an urgent cybersecurity threat with a compressed timeline of months rather than years before significant impact. This represents an unusual coordinated government warning about AI-enabled cyber operations.
CVE-2026-20230, a server-side request forgery flaw in Cisco Unified Communications Manager, is now actively exploited following public proof-of-concept availability since June. Organisations running affected UCM instances face network reconnaissance and potential lateral movement risks.
OpenDJ's JMX RMI connector deserializes untrusted data before authentication, enabling unauthenticated RCE via malicious serialized objects. This PoC validates a critical pre-auth bypass affecting all versions through 5.1.0.
scim-patch ≤0.9.0 allows attackers to pollute Object.prototype through crafted SCIM PATCH requests, enabling privilege escalation and logic bypass across entire Node processes that process untrusted IdP input.
Unit 42 has disclosed a bucket hijacking technique that exploits global namespace uniqueness across cloud storage systems, allowing attackers to intercept and exfiltrate data by claiming abandoned or predictably-named buckets before legitimate organisations do.
Squidbleed is a memory disclosure vulnerability in Squid proxy that can leak sensitive user data from process memory, affecting organisations using Squid for HTTP caching and filtering. The flaw's age and ubiquity in enterprise networks creates significant exposure.
OpenAI is expanding its Daybreak initiative by releasing an improved GPT-5.5-Cyber model to trusted defenders for identifying and patching software vulnerabilities across large codebases. This represents a maturing capability in AI-assisted security testing that could reshape how organisations approach vulnerability discovery.
Brazil's Civil Defense Alert system was breached and weaponised to distribute at least a dozen unauthorised emergency alerts, demonstrating how compromised critical infrastructure can erode public trust in legitimate disaster warnings.
Attackers are conducting a multi-country phishing campaign via WhatsApp that distributes malicious VBScript files disguised as business documents, enabling remote code execution and system compromise on Windows PCs.
Hackers breached a third-party license vendor serving Texas Parks & Wildlife Department, stealing personal data for approximately 3 million individuals. The incident highlights the expanded attack surface organisations face when outsourcing customer-facing systems.
AryStinger, a previously undocumented botnet, has compromised over 4,000 outdated D-Link routers worldwide to operate as proxy infrastructure for malicious traffic. The campaign highlights the security debt posed by end-of-life networking equipment still deployed in production environments.
Microsoft attributed a compromise of over 140 npm packages to North Korean threat actor Sapphire Sleet, leveraging the Mastra AI project as an entry point. This represents a high-impact supply chain attack with potential for widespread malware distribution across the JavaScript ecosystem.
French President Macron has called for wealthy democracies to coordinate on AI regulation and urged the US to share advanced AI capabilities. This represents a significant geopolitical positioning around AI governance standards.
Attackers are exploiting an unauthenticated information disclosure flaw in Gravity SMTP (CVE-2026-4020) to extract API keys, OAuth tokens, and configuration secrets from roughly 100,000 WordPress installations. The vulnerability remains exploitable despite patches because many sites run unpatched versions.
A new ransomware family named Prinz Eugen prioritises recently modified files for encryption and omits ransom notes, suggesting operators are optimising for speed and evasion. This behaviour pattern indicates a maturing threat actor focused on minimising forensic artefacts.
An Insecure Direct Object Reference (IDOR) vulnerability in Langflow's `/api/v1/responses` endpoint allows authenticated attackers to access and execute flows owned by other users by manipulating flow UUIDs. The vulnerability bypasses user ownership validation when flows are referenced by UUID rather than endpoint name.
Path traversal via symlink injection in tar extraction allows unauthenticated file disclosure, JWT secret theft, and remote code execution in Langflow RAG deployments. Organizations must patch immediately as exploit chain is straightforward and affects multiple file-handling components.