< ciso
brief />
Tag Banner

All news with #ai security tag

1155 articles · page 4 of 58

Google unveils Gemini 4 Carbon with cybersecurity lead

🛡️ Google announced Gemini 4 Carbon, a new frontier model focused on advanced cybersecurity and long-horizon capabilities, claiming leading scores on benchmarks like SEC-bench Pro, ExploitGym, and CWE-bench. The company highlighted Carbon's improvements in vulnerability detection, exploitation validation, and patching, and said it is more likely to seek human input under ambiguity and resist indirect prompt-injection attacks. The rollout mirrors Gemini 4 Argon, initially limited to select cyber defenders under the Fairwind Program while Google engages in U.S. pre-release testing.
read more →

Designing Cyber-Immune AI Agents for Secure Rollout

🛡️ Kaspersky experts outline how adopting Cyber Immunity and Security by Design principles helps reduce risks when moving from simple assistants to autonomous AI agents. The article explains why AI systems are harder to secure — probabilistic outputs and blurred boundaries between commands and data — and highlights common pitfalls such as sandbox escapes, configuration attack surfaces, and user fatigue from frequent prompts. It recommends limiting capabilities, minimizing the trusted computing base, maintaining inventories, and increasing observability to log model requests, responses, and reasoning.
read more →

Audio Guardrails for Securing Voice AI Agents

🔊 Audio guardrails, part of Check Point AI Security, screen the raw audio a voice agent receives to detect prompt attacks before the model acts. Voice agents increasingly bypass transcripts by using speech-to-speech models, creating opportunities for audio prompt injection and machine-focused manipulations that text-only checks miss. The system runs two ordered checks: an audio-integrity check that detects degradation or engineered noise and a transcript-based prompt-attack scan in over 80 languages. Verdicts return to the application so it can block, request repetition, or escalate to a human.
read more →

The Rise of the Threat Operator in AI-driven SOCs

🛡️ AI is beginning to take over many traditional security analyst tasks, from investigating alerts to executing response actions, prompting a reorganization of SOC work and decision-making. As AI handles operational workloads, the emerging Threat Operator role will direct AI-driven defense, translate business risk into priorities, and judge automated decisions. The shift requires new training, governance, and career paths that combine technical fundamentals with oversight and risk management.
read more →

Online Scam Volumes Double, Overtaking Phishing

📈 Akamai’s State of the Internet report shows online scam detections doubled between July 2025 and June 2026, with scams overtaking phishing for the first time. The vendor recorded about 6.4 million scams in April–June 2026 versus 3.2 million in July–September 2025, including fake stores, investment and tech support scams. Phishing also rose by 61%, driven by reusable kits, industrialized infrastructure and AI, which shortened average attack lifespans. Security teams increasingly use AI to accelerate investigations and disrupt attacks faster.
read more →

Securing MCP Servers Ahead of AI Rollouts

🛡️ This article explains the role of Model Context Protocol (MCP) servers in LLM deployments and emphasizes the need to secure them before widespread AI rollout. It outlines top-priority controls such as input/output validation, network segmentation, server hardening, least-access principles, encryption, and tool management. The piece also recommends third-party assessments, robust logging, human-in-the-loop for high-risk actions, and pre-production testing to confirm controls are effective. Finally, it advocates using a risk management approach and stakeholder engagement to prioritize implementation.
read more →

Wazuh Integrates AI to Enhance SOC Workflows

🛡️ Wazuh introduces AI-powered capabilities to augment Security Operations Centers (SOCs), offering the Wazuh AI Analyst on Wazuh Cloud and integrations with third-party and self-hosted models. The service delivers automated, scheduled reports and in-console insights while ensuring subscription data is not used for model training. Organizations can choose self-hosted LLMs like Llama 3 via Ollama for privacy or cloud-hosted models such as Anthropic’s Claude on Amazon Bedrock for broader capabilities.
read more →

Anthropic Model Chains Found and Chained Multiple Bugs

🔍 Anthropic reported that an unreleased internal AI model discovered security flaws across Chrome, Safari, Android, Windows, and Linux and chained separate bugs into functioning proof-of-concept exploits. The company says the exploits were reproduced in the lab after prompting the model, were not observed in the wild, and were disclosed to vendors before publication; fixes are already available. Anthropic described five distinct chains and credits for several fixes reference the company and its Claude model.
read more →

AI-driven campaigns target orgs using multi-agent tools

🛡️ TrendAI and Trend Micro researchers detailed three cyber campaigns in 2026 where AI played central or supportive roles. The incidents ranged from a fully automated CyberStrikeAI operation using 44 agents to human-led attacks aided by large language models and AI-assisted tool development. Targets included VPNs, WebLogic, GIS servers and edge devices across Japan and multiple other countries. The analysis highlights faster weaponization, automated reconnaissance, and commoditized offensive AI tools.
read more →

ISC2 launches Cyber‑AI Workforce Coalition

🔎 ISC2 has launched the Cyber‑AI Workforce Coalition to help shape the evolving AI security workforce. Announced October 20 and introduced by CEO Scott Beale at ISC2 Security Congress 2026, the coalition will act as a vendor‑neutral convener of employers, standards bodies and practitioners. It aims to produce practical toolkits — not new standards — to help employers write roles, hire and upskill staff for AI security needs. The group’s first meeting is in November, with toolkits expected by spring 2027.
read more →

ISC2 to launch first AI security certification

🛡️ ISC2 announced a new AI-focused security certification at Security Congress 2026 in Denver, with a pilot exam opening October 26 and a beta planned for December 2026. The credential, still unnamed, targets professionals securing AI-enabled environments and was shaped by input from 10,000 volunteers. ISC2 seeks 3,000 beta volunteers by November 4 and aims to make the certification available in early 2027. Eligibility requires relevant cybersecurity/AI experience, with some ISC2 credentials substituting experience requirements.
read more →

AI widens blast radius: breached data fuels follow-on attacks

🔍 Lab 1’s 2026 report finds the median breach now exposes 542 organizations beyond the initial victim, as AI accelerates analysis of stolen datasets. The median breach size climbed to 29.8GB and 32,734 files, with 44% containing financial data and 9% containing HR records; private cryptographic keys appeared in ~15% of incidents. Criminals and specialist groups now offer indexing-as-a-service to mine and monetize breaches, increasing downstream fraud, extortion, and targeted attacks. CISOs must track where sensitive data sits across supply chains and demand rapid notification from third parties.
read more →

Extending Network Security to AI Applications

🔒 FortiAIGate extends traditional network and application security by inspecting AI prompts, responses, and tool interactions. It provides a centralized, self-hosted AI gateway that enforces guardrails, data protection, and usage controls for LLMs, agents, and MCP tools. Deployable on-premises or in cloud Kubernetes clusters, it integrates with FortiGate to detect and manage shadow AI activity and sensitive data exposures.
read more →

EvilTokens and defending against AI-assisted attacks

🔐 Cybercriminals are integrating AI into traditional attacks, accelerating reconnaissance and post-compromise activity. EvilTokens, a phishing-as-a-service targeting Microsoft 365 via the OAuth device authorization flow, uses AI to analyze environments and craft convincing follow-up frauds. Security metrics like MTTD and MTTR must be re-evaluated, and defenses should be tested end-to-end to understand how controls perform together. Preemptive security—identifying plausible attacker actions before they occur—is increasingly important.
read more →

The 10 Most Powerful Cybersecurity Companies

🔒 This analysis examines the top 10 cybersecurity companies adapting to an AI-driven threat landscape, highlighting market leaders and fast-growing challengers. It details major vendors’ strategic AI integrations, recent acquisitions, flagship products, and key financial metrics. The piece emphasizes how incumbents like Palo Alto, Microsoft, Cisco, and Google are embedding AI across platforms to protect agentic workloads and data. It also notes smaller innovators gaining traction and reshaping enterprise security priorities.
read more →

Microsoft Teams to add AI context-aware phishing detection

🔒 Microsoft will deploy AI-powered, context-aware phishing detection in Teams starting early November 2026, with a targeted release finishing by mid-November and general availability expected between early and mid-December. The feature analyzes conversation context to detect phishing, impersonation, and deceptive intent in higher-risk scenarios like external first-contact chats, without retaining message content for AI training. IT admins do not need to enable the protection, which focuses on external first-contact chats for most users.
read more →

AI-driven intrusion ecosystem linked to ransomware groups

🛡️ Cybersecurity researchers uncovered an AI-driven intrusion ecosystem run by affiliates of the INC and The Gentlemen ransomware groups. The operation centers on an "Autonomous Red Team Kill Chain Framework" named Phantom and leverages exposed edge and remote-access systems at scale. Operators used AI-managed knowledge bases, OpenCode with OpenRouter-hosted models, and a human-approved workflow to execute intrusions, lateral movement, persistence, and data exfiltration. SOCRadar identified numerous exploited vulnerabilities and a toolkit including Ghost C2, RevSocks, WormGPT-MCP, and bespoke EDR evasion techniques.
read more →

Palo Alto Networks Adds AI Defense to OpenAI Marketplace

🔒 Palo Alto Networks has made Unit 42 Continuous Frontier AI Defense and Prisma AIRS available through the OpenAI Marketplace, enabling OpenAI enterprise customers to procure proactive, machine-speed security alongside their AI spend commitments. The announcement responds to accelerating attacks where AI compresses breach cycles dramatically, and offers organizations an option to convert OpenAI commitments into continuous defense subscriptions. These solutions combine offensive security expertise, a proprietary multi-model harness using defender-grade models, and centralized AI security governance.
read more →

After the Mythos Moment: Hard Work for Cybersecurity

🔐 Anthropic’s Claude Mythos Preview provoked alarm by autonomously finding and exploiting vulnerabilities, accelerating tasks once reserved for expert pen testers. The initial shock of limited preview access has faded, and organizations now face the harder task of turning that warning into practical defensive readiness. Defenders must adopt AI-driven exposure management, prioritize real risks and automate patching to keep pace with attackers operating at machine speed.
read more →

AI Notetakers: Ubiquitous Recording Risks and Controls

📝 AI-driven meeting notetakers are rapidly moving off-platform and into wearable and ambient devices, creating pervasive recording risks. Vendors and employees often adopt these tools to improve efficiency, but shadow usage, weak vendor security, and inconsistent legal frameworks increase exposure. Organizations must adopt clear disclosure, classification, retention, and vendor controls to manage consent, litigation, and data-leak risks.
read more →