All news with #ai security tag

🛡️ A malware developer has begun embedding provocative text about nuclear and biological weapons inside large JavaScript block comments in spyware payloads to confuse AI-based scanners. The commented header is ignored at runtime but aims to trigger refusals or misclassification in naive LLM-powered triage systems that ingest file starts without isolating untrusted content. Traditional detection methods—YARA, entropy checks, AST parsing, and behavioral analysis—remain effective, but the technique is a practical anti-analysis tactic against weak AI-first pipelines.

🛡️ A ReliaQuest report finds AI is making cyber-attacks cheaper, faster to scale, easier to customize and harder to spot while not fundamentally altering attacker tradecraft. Initially used for polishing phishing and basic scripting in 2024, by mid-2025 AI had expanded into deepfakes, AI-assisted scripts and an underground market for tools. Today AI appears embedded in workflows—generating phishing pages, web shells, and obfuscating code—and as the lure itself, with attackers leveraging trusted AI brands to trick users.

🔒 This article examines the rising need for AI security posture management (AI-SPM) as enterprises adopt AI across workflows. It outlines how AI maturity stages — from AI-assisted to AI-native — change security requirements and why agents and model services expand the attack surface. The piece surveys vendor approaches, key features, and integrations, and provides guidance for selecting AI-SPM solutions to avoid coverage gaps.

🛑 Meta has frozen its Model Compatibility Initiative (MCI) after employees reportedly bypassed guardrails and accessed sensitive internal data, then did so again after an attempted fix. The program collected inputs like keystrokes, mouse movements, clicks, and screen content to train AI, and employees were initially not allowed to opt out. Meta says it found unauthorized access on June 18 and paused MCI while investigating, asserting no indication yet of improper access beyond what was reported. Analysts criticized inadequate protections and insufficient risk tagging for highly sensitive non-PII telemetry.

🛡️ AWS previewed AI-powered investigations in Amazon GuardDuty to automate analysis of findings and reduce manual investigation time. The capability uses knowledge graphs and threat intelligence to examine 90 days of related activity, affected resources, and indicators, delivering disposition assessments with confidence scores, MITRE ATT&CK classifications, evidence, and remediation recommendations. Available in preview in 10 regions and accessible via the GuardDuty console, CLI, API, or AWS' MCP Server.

🔒 Google Cloud announces expanded Confidential Computing capabilities to protect data in use for AI workloads. The update includes Confidential G4 VMs with NVIDIA RTX PRO 6000 Blackwell GPUs, open-source Prompt Encryption SDKs, Intel TDX on C4 machines, and broader Confidential Space enhancements. These innovations aim to provide verifiable attestation, end-to-end encrypted inference, and support for multi-party collaboration while preserving performance and scalability.

🔒 OpenAI has expanded its Daybreak cyber-defense program, advancing patch automation with the full release of GPT-5.5-Cyber, updates to Codex Security, and a new open-source patching initiative. Access to the model is limited to verified defenders and paired with enhanced monitoring. OpenAI reports improved vulnerability reproduction and exploit-writing scores, while emphasizing human oversight and partnerships with vendors and governments.

🔒 Offensive AI is shifting tools from drafting to autonomous action. Agentic systems can gather intelligence, craft tailored social engineering, and run exploit chains without human hands, expanding capability to unskilled actors while accelerating expert operations. Defenders must test protections with live adversarial use to understand real resilience and retain human judgment where agents remain prone to confident errors.

🛡️ The Five Eyes cybersecurity agencies warned on June 22 that frontier AI is already reshaping offensive and defensive cyber capabilities and urged businesses to prioritize cyber resilience. They cautioned that AI accelerates attacks by lowering barriers and shrinking the window between discovery and exploitation, while also offering defensive benefits. The group recommended a whole-of-organization response focused on basics, secure-by-design, defence in depth, and integrating AI into security operations. Practical steps included reducing attack surfaces, accelerating patching, addressing legacy systems, strengthening access controls, and preparing incident response.

🔒 OpenAI is distributing an enhanced GPT‑5.5‑Cyber model to trusted defenders via the Daybreak program, claiming improved capability to find, validate, and patch software vulnerabilities across large codebases. The company also updated the Codex Security plugin to accelerate discovery, triage, and automated patch generation, and launched Patch the Planet with Trail of Bits to secure open‑source projects. These steps aim to help maintainers cope with the surge in AI‑driven vulnerability findings while preserving human oversight.

🔒 At Zscaler’s Zenith Live 2026 in Vienna, the vendor argued that AI agents are rapidly becoming digital workers while regulators tighten data residency and supply‑chain threats move closer to core operations. Zscaler proposes extending its Zero Trust Exchange and SASE platform to govern AI agents, unmanaged devices, multi‑cloud workloads, and B2B partners, positioning zero trust as the control plane for secure AI adoption in regulated, highly connected markets like Southeast Asia. The company emphasised an AI Broker, endpoint AI security, and an AI Access Graph to map and protect AI assets and data flows.

🛡️ Shadow AI has evolved from simple data leakage to an access control challenge as employee-built agents connect to enterprise systems. These agents — created across platforms, extensions, and scripts — can call APIs, use credentials, and perform actions in production, often with broad or forgotten permissions. Traditional controls like DLP and domain blocking miss non-human identities, so organizations must inventory agents, map ownership and credentials, and enforce automated remediation.

🔍 A simple framework called the SOC Triangle balances quality, consistency and cost efficiency in security operations. Human-centric workflows create trade-offs where improving one dimension often harms another. AI is changing this dynamic by automating repeatable investigative workflows, improving depth, consistency and scaling without linear headcount increases. The triangle still exists, but its constraints are loosening for machine-suitable tasks, shifting humans toward oversight and complex judgment.

🔒 A proof-of-concept called SearchLeak demonstrated a prompt injection attack against Microsoft M365 Copilot Enterprise that tricks users into clicking crafted links to exfiltrate corporate data. Researchers combined three weaknesses in Copilot Search — including URL query parameters treated as natural language prompts — to leak sensitive content. Microsoft patched the server-side flaw, but the incident highlights risks when AI services access broad corporate assets and the need for render-time sanitization and stricter CSPs.

🚀 Palo Alto Networks is launching a local cloud location for Prisma® AIRS™ in Japan to secure emerging AI deployments. The expansion provides domestic data residency, low-latency processing, and phased rollout of comprehensive AI security features for models, agents, and artifacts. It aims to help Japanese organizations adopt Generative AI and agentic workflows with improved operational efficiency and cyber resilience.

🛰️ Cisco Talos' Threat Source newsletter reflects on how human behavior, context, and competing priorities often override rational security decisions. The piece links a Spielberg film theme to cybersecurity, noting that knowledge alone doesn't ensure action — organizations struggle with budgets, workloads, and urgency. Talos highlights practical controls like segmentation, backups, and MFA, and showcases a new reverse-engineering method that pairs local AI agents with tools like vbdec to accelerate analysis while protecting sensitive binaries.

🔧 This post describes how Cloudflare evolved a single-repo security skill into a fleet-scale, model-agnostic Vulnerability Discovery Harness (VDH) and a separate Vulnerability Validation System (VVS). It explains why single-agent prompts fail at scale and why treating models as interchangeable components improves coverage. The article outlines stages like Recon, Hunt, Validate, Trace, Dedup, Gapfill, and Feedback and emphasizes persistence, strict context controls, and cross-repo reasoning.

📰 This week's ThreatsDay roundup highlights a range of active campaigns and emerging risks, from DoH adoption in Windows Server 2025 to search-hijacking Chrome extensions and fileless macOS infections. Researchers uncovered abuse of shared AI chat features to deliver credential stealers, large-scale WhatsApp booking fraud, and memory-only stealers targeting banks. Vendors and agencies are responding with mitigations, advisories, and new product timelines to address quantum and AI-driven threats.

🔎 Sophos CTU research finds cybercriminals debating the risks and benefits of AI tools across underground forums, marketplaces and messaging apps. Sellers are offering AI kits for phishing, malware automation, deepfake creation and social engineering, while some threat actors fear losing work to automated toolsets. The research highlights divided views, a spike in discussion after the release of Claude Mythos Preview, and advice for defenders to prioritize patching, MFA and visibility.

🛡️ At least one malware author is inserting large comment blocks with policy-triggering content about nuclear and biological weapons into JavaScript payloads to disrupt AI-driven analysis. The decoy text sits inside comments so execution is unchanged while early-stage LLM-based triage can be confused or refuse to process the file. Traditional detection methods like YARA rules, entropy checks, and deobfuscation remain effective. This tactic targets naive pipelines that expose untrusted file starts to language models.