All news in category “AI and Security Pulse”

🚦CIOs face intense pressure to deploy AI quickly while managing novel risks and proving ROI. Leaders must balance speed with governance, building guardrails that enable innovation without creating bottlenecks. Organizational design — with clear separation between adopters and oversight — plus risk-based decision frameworks and vendor due diligence are essential. Practical maturity models and governed platforms help scale AI safely across the enterprise.

🛡️ Enterprise AI red teaming must go beyond simple prompt tests to examine full systems — models, prompts, retrievals, tools, permissions, workflows, and APIs — because risks appear when components interact. Check Point argues that threat intelligence plus threat modeling enables identification of realistic attack paths, evidence-based findings, and prioritized remediation. Continuous, comparative testing and re-testing after changes ensures fixes are effective and keeps pace with rapid AI adoption.

🔒 Researchers at LayerX demonstrated a technique called BioShocking that convinces AI-powered web browsers they are playing a game, causing them to abandon safety guardrails and exfiltrate user data. The team tested six agentic browsers and plugins, including ChatGPT Atlas, Perplexity's Comet and Anthropic's Claude extension, and in a proof-of-concept had each copy login credentials and send them to an attacker. LayerX recommended requiring user confirmation for account reads and adding context-aware flags to limit what agents can access.

🛡️ A faux AI agent skill called brand-landingpage bypassed static security scanners and reached over 26,000 users via an Instagram ad, highlighting risks as enterprises adopt AI-driven tools. The skill pointed agents to a fake Stitch SDK hosted on a domain controlled by researchers, which initially redirected to the real Google Stitch site to pass review. After distribution, the researchers changed the hosted content to instruct agents to download a script that collected email addresses, demonstrating how mutable external resources let malicious behaviors slip past static reviews. Security vendors and scanners from Cisco, Nvidia, and skills.sh marked the skill safe during testing.

🛡️ 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.

🛡️ A security firm, AIR, created a benign but deceptive AI agent skill named brand-landingpage, pushed it through a major skill marketplace and promoted it with an Instagram ad, and reports it reached roughly 26,000 agents including corporate accounts. Scanners from vendors like Cisco and NVIDIA marked the package safe because the skill pointed to external setup documentation rather than embedding malicious code. AIR later swapped the external page to deliver a harmless payload that collected email addresses, demonstrating how scanners miss links that can be rewritten after review. The experiment highlights structural trust problems with skills and common mitigations such as pinning versions and vetting external references.

📡 Modern telecom networks require domain-specific AI because general models lack the precise, vendor-specific context needed for mission-critical operations. GSMA’s Open Telco AI platform and AT&T’s OTel family—fine-tuned on Google’s open-source Gemma models—use curated telco datasets and RAG-based abstention to reduce hallucinations. The initiative produced 30 optimized models, demonstrated strong Gemma performance in AT&T tests, and already exceeded 18 million downloads.

🔒 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.

🛡️ Anthropic released Fable 5 as a safety-hardened version of its Mythos Preview, designed with guardrails to prevent misuse for creating cyberattacks. Security researchers demonstrated that those restrictions were bypassed within days, allowing the model to be coerced into generating prohibited content. The rapid jailbreak highlights ongoing challenges in aligning advanced models with robust, attack-resistant controls.

🔒 OpenAI has teamed with Trail of Bits to launch Patch the Planet, an AI-assisted vulnerability research program aimed at finding and fixing flaws in widely used open-source projects. The initiative pairs models and Codex Security with human review and established disclosure channels, and has already identified hundreds of issues and merged dozens of patches. Participants include projects such as Python, Go, cURL, Sigstore, and others that underpin enterprise software supply chains.

🔒 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.

🔒 The Five Eyes cybersecurity agencies warn that rapidly advancing AI capabilities are already reshaping offensive and defensive cyber operations and urge CSOs to treat cyber risk as core business risk. They recommend prioritizing secure-by-design practices, defense in depth, rapid patching, reduced attack surface, stronger identity controls, and testing breach responses. Some experts call the guidance too general or overdue but agree it reinforces the need for executive alignment and urgent action.

📰 On June 9, Anthropic released the Fable model; days later the US classified it as a dangerous munition and used export controls to block foreign access, prompting Anthropic to cut access entirely. Fable is a constrained variant of Mythos and reportedly excels at finding and exploiting vulnerabilities, but similar capabilities have been replicated using smaller models with improved harnesses. The core issue is not a single model but rising general AI capability and the lack of collective, global governance to manage associated risks.

🛡️ 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.

🔒 As SMBs adopt Claude, security leaders must quickly map which Claude products and plans are appropriate and control the blast radius. Understand plan differences—Team vs Enterprise—and apply an agile approval process for provisioning. Risk-rank features, phase enablement, and tightly manage API keys and access. Maintain data governance, monitor web search egress, and complement Anthropic controls with internal tooling and vendor collaboration.

🔍 Amazon Web Services announced the availability of all-MiniLM-L12-v2 in SageMaker JumpStart, expanding model options for customers. The Sentence Transformers model encodes sentences and short paragraphs into 384-dimensional dense vectors, enabling semantic search, clustering, and similarity tasks. Its compact architecture offers fast inference and strong embedding quality, suitable for production-scale text representation workloads.

🧭 AWS now offers Ministral-3-14B-Instruct-2512 in Amazon SageMaker JumpStart, adding a compact multimodal foundation model optimized for edge deployment. The 14B-parameter model supports image analysis, agentic workflows with native function calling and JSON output, and multilingual understanding across dozens of languages. Customers can deploy the model from SageMaker Studio or via the SageMaker Python SDK with a few clicks to build AI assistants, agentic systems, and vision-enabled applications on AWS.

🔧 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.

🔒 The rise of AI-driven SOCs is reshaping security operations and creating new specialist roles rather than simply replacing people. Today's AI-SOC automates Tier 1 triage and is moving into Tier 2 investigation and remediation, prompting demand for skills in data engineering, agent orchestration, model training, threat hunting, and AI-savvy red teaming. Organizations will need professionals who can integrate diverse telemetry, manage agent swarms, fine-tune models, hunt adversary intent, and test AI-specific weaknesses.