All news with #agent security tag

🔒 Microsoft argues that as organizations adopt agentic AI, security must be a strategic priority that enables growth, trust, and continued innovation. The post identifies risks such as oversharing, data leakage, compliance gaps, and agent sprawl, and recommends three pillars: prepare for AI and agent integration, strengthen organization-wide skilling, and foster a security-first culture. It points to resources like Microsoft’s AI adoption model, Microsoft Learn, and the AI Skills Navigator to help operationalize these steps.

🛡️ Google announced a set of layered security measures for Chrome after adding agentic AI features, aimed at reducing the risk of indirect prompt injections and cross-origin data exfiltration. The centerpiece is a User Alignment Critic, a separate model that reviews and can veto proposed agent actions using only action metadata to avoid being poisoned by malicious page content. Chrome also enforces Agent Origin Sets via a gating function that classifies task-relevant origins into read-only and read-writable sets, requires gating approval before adding new origins, and pairs these controls with a prompt-injection classifier, Safe Browsing, on-device scam detection, user work logs, and explicit approval prompts for sensitive actions.

🔐 This post from Adrien Delaroche at Google Cloud outlines three architectures for AI agents that interact with blockchains: the agent-controlled custodial model, a self-hosted variant, and the non-custodial transaction-crafter model. It explains security, performance, and malice risks when agents hold private keys and recommends returning unsigned transactions so users sign locally. The author demonstrates a sample implementation using Google ADK, Gemini 2.0 Flash, Cloud Run, and an Ethereum faucet, and urges MCP servers to support both signing and unsigned flows to balance automation with user safety.

🚨 The emergence of agentic AI browsers converts the browser from a passive viewer into an autonomous digital agent that can act on users' behalf. To perform tasks—booking travel, filling forms, executing payments—these agents must hold session cookies, saved credentials, and payment data, creating an unprecedented attack surface. The piece cites OpenAI's ChatGPT Atlas as an example and warns that prompt injection and the resulting authenticated exfiltration can bypass conventional MFA and network controls. Recommended mitigations include auditing endpoints for shadow AI browsers, enforcing allow/block lists for sensitive resources, and augmenting native protections with third-party browser security and anti-phishing layers.

🤖 Agentic AI is emerging as a practical accelerator for security teams, automating detection, triage, remediation and routine operations to improve speed and scale. Security leaders at Zoom, Dell, Palo Alto and others highlight its ability to reduce alert fatigue, augment SOCs and act as a force multiplier amid persistent skills shortages. Implementations emphasize augmentation over replacement, enabling continuous monitoring and faster, more consistent responses.

🔎 Amazon Web Services has introduced Agentic Search for OpenSearch Service, an agent-driven layer that interprets natural-language intent, orchestrates search tools, and generates OpenSearch DSL queries while providing transparent summaries of its decision process. The built-in QueryPlanningTool uses LLMs to plan and emit DSL, removing the need for manual query syntax. Two agent types are available: conversational agents with memory and flow agents optimized for throughput. Administrators can configure agents via APIs or OpenSearch Dashboards, and Agentic Search is supported on OpenSearch Service version 3.3+ across AWS Commercial and GovCloud regions.

🚀 Claude Opus 4.5 is now available in public preview in Microsoft Foundry, aiming to shift models from assistants to agentic collaborators that execute multi-tool workflows and support complex engineering tasks. Anthropic and Microsoft highlight Opus 4.5’s strengthened coding, vision, and reasoning capabilities alongside improved safety and prompt-injection robustness. Foundry adds developer features like Programmatic Tool Calling, Tool Search, Effort Parameter (Beta), and Compaction Control to help teams build deterministic, long-running agents while keeping centralized governance and observability.

🤖 AWS introduces the Agentic AI Security Scoping Matrix to help organizations secure autonomous, tool-enabled AI agents. The framework defines four architectural scopes—from no agency to full agency—and maps escalating security controls across six dimensions, including identity, data/memory, auditability, agent controls, policy perimeters, and orchestration. It advocates progressive deployment, layered defenses, continuous monitoring, and retained human oversight to mitigate risks as autonomy increases.

⚠️ In mid‑September 2025, Anthropic detected a sophisticated espionage campaign in which attackers manipulated its Claude Code tool to autonomously attempt infiltration of roughly thirty global targets, succeeding in a small number of cases. The company assesses with high confidence that a Chinese state‑sponsored group conducted the operation against large technology firms, financial institutions, chemical manufacturers, and government agencies. Anthropic characterizes this as likely the first documented large‑scale cyberattack executed with minimal human intervention, enabled by models' increased intelligence, agentic autonomy, and access to external tools.

🤖Agentic AI gives autonomous agents the ability to access external systems, gather information, and take actions within defined workflows, making routine multi-system tasks far more efficient for human operators. Cisco Talos warns this efficiency is already being mirrored in the cyber crime economy, including the first observed AI-orchestrated campaign in early 2025. While AI lowers barriers to entry and speeds operations for attackers, it is imperfect and still requires skilled instruction and human oversight. Defenders can respond by building their own agentic tools, deploying honeypots to engage malicious agents, and refining detection to stay ahead.

📊 Google introduces BigQuery Agent Analytics, an ADK plugin that streams agent interaction events into BigQuery to capture, analyze, and visualize performance, usage, and cost. The plugin provides a predefined schema and uses the BigQuery Storage Write API for low-latency, high-throughput streaming of requests, responses, and tool calls. Developers can filter and preprocess events (for example, redaction) and build dashboards in Looker Studio or Grafana while leveraging vector search and generative AI functions for deeper analysis.

🚀 At Microsoft Ignite 2025, Microsoft unveiled a suite of Azure infrastructure and AI operational innovations built for scale, reliability, and security. Azure Copilot introduces an agentic operations model with six specialized agents—migration, deployment, optimization, observability, resiliency, and troubleshooting—designed to automate routine cloud management while enforcing RBAC and policy. The release also highlights new AI datacenter architecture (Fairwater), deployment of NVIDIA GB300 GPUs at scale, and platform improvements like Azure Boost and AKS Automatic to accelerate performance and reduce operational overhead.

🛡️ At Microsoft Ignite 2025, Microsoft set out an ambient, autonomous security approach for the emerging agentic era and announced a suite of tools to observe, secure, and govern AI agents and apps. The centerpiece is Microsoft Agent 365, a control plane providing an Entra-based registry, access controls, visualization, and integrations with Defender, Entra, and Purview to detect prompt-injection, prevent leakage, and enable auditing. Microsoft also expanded platform protections, enhanced Copilot data controls in Purview, and positioned Microsoft Sentinel and Security Copilot as agentic security pillars for detection and response.

🧭 Hugo Selbie presents a practical framework for evaluating modern multi-step AI agents, emphasizing that final-output metrics alone miss silent failures arising from incorrect reasoning or tool use. He recommends defining clear, measurable success criteria up front and assessing agents across three pillars: end-to-end quality, process/trajectory analysis, and trust & safety. The piece outlines mixed evaluation methods—human review, LLM-as-a-judge, programmatic checks, and adversarial testing—and prescribes operationalizing these checks in CI/CD with production monitoring and feedback loops.

🤖 This recap of The Agent Factory episode summarizes a conversation between Amit Maraj and Ravin Kumar (DeepMind) about building open-source agentic models. It highlights how agent training differs from standard ML, emphasizing trajectory-based data, a two-stage approach of supervised fine-tuning followed by reinforcement learning, and the paramount role of evaluation. Practical guidance includes defining a 50-example final exam up front and considering hybrid setups that use a powerful API like Gemini as a router alongside specialized open models.

🔐 As enterprises deploy AI assistants and autonomous agents, existing security frameworks must evolve to treat these agents as first-class identities rather than afterthoughts. The piece advocates applying Zero Trust principles—identity-first access, least-privilege, dynamic contextual enforcement, and continuous monitoring—to agentic identities to prevent misuse and reduce attack surface. Practical controls include scoped, short-lived tokens, tiered trust models, strict access boundaries, and assigning clear human ownership to each agent.

🚀 Google details a broad set of GKE and Kubernetes enhancements announced at KubeCon to address agentic AI, large-scale training, and latency-sensitive inference. GKE introduces Agent Sandbox (gVisor-based) for isolated agent execution and a managed GKE Agent Sandbox with snapshots and optimized compute. The platform also delivers faster autoscaling through Autopilot compute classes, Buffers API, and container image streaming, while inference is accelerated by GKE Inference Gateway, Pod Snapshots, and Inference Quickstart.

🛡️ Agent Sandbox is a new Kubernetes primitive designed to run AI agents with strong, kernel-level isolation. Built on gVisor with optional Kata Containers and developed in the Kubernetes community as a CNCF project, it reduces risks from agent-executed code. On GKE, managed gVisor, container-optimized compute and pre-warmed sandbox pools deliver sub-second startup latency and up to 90% cold-start improvement. A Python SDK and a simple API abstract YAML so AI engineers can manage sandbox lifecycles without deep infrastructure expertise; Agent Sandbox is open source and deployable on GKE today.

🧭 This post explains when to use sub-agents versus packaging agents as tools when building multi-agent systems with Google's Agent Development Kit (ADK). It contrasts agents-as-tools — encapsulated, stateless specialists invoked like deterministic function calls — with sub-agents, which are stateful, context-aware delegates that manage multi-step workflows. The guidance highlights trade-offs across task complexity, context sharing, reusability, and autonomy, and illustrates the patterns with data-agent and travel-planner examples to help architects choose efficient, scalable designs.

🔐 Computer-using agents (CUAs) — AI systems that perceive screens and act like humans — are poised to scale phishing and credential-stuffing attacks by automating UI interactions, adapting to layout changes, and bypassing anti-bot defenses. Organizations should move beyond passwords and shared-secret MFA to device-bound, cryptographic authentication such as FIDO2 passkeys and PKI-based certificates to reduce large-scale compromise. SaaS vendors must integrate with identity platforms that support phishing-resistant credentials to strengthen overall security.