All news with #agent security tag

🔒 Palo Alto Networks and Databricks announce an integrated runtime security solution to protect agentic AI across the enterprise. The partnership embeds Prisma AIRS into the Databricks Unity AI Gateway to provide centralized governance, real-time inspection, and policy-driven enforcement for prompts, tool calls, and model interactions. This approach aims to prevent prompt injections, data exfiltration, and malicious tool usage while enabling faster, secure AI deployments.

🤖 Google announces expanded Agentic Data Cloud capabilities, introducing new data agents and tools to enable conversational analytics and agent-driven workflows across BigQuery, Lakehouse, AlloyDB, Spanner, and Cloud SQL. The update includes Data Engineering, Data Science, Database Observability, Looker Dashboard, Data Insights, and Deep Research agents, plus developer toolkits like the Data Agent Kit and Managed MCP servers. These features aim to ground agents in real-time enterprise data with unified governance and near-100% accuracy for tasks such as NL-to-SQL conversions and automated pipeline maintenance.

🧭 This post describes how Yahoo and Google Cloud built Seller Agent, an agentic media-buying platform that collapses multi-week manual workflows into governed campaigns executed in seconds. The architecture uses a dual-graph approach — a knowledge graph for deterministic business logic and a context graph for auditable decision traces — combined with Google Cloud services like Spanner Graph, BigQuery Graph, and Gemini. The design emphasizes explainability, regulator-grade governance, and closed-loop learning to ensure autonomous actions remain transparent and accountable.

🛡️ New research shows that reasoning-based AI agent guardrails can be weaponized into denial-of-service vectors by a single poisoned document that traps safety systems in extended thinking loops. The study, from the Hong Kong University of Science and Technology and collaborators, demonstrated large slowdowns across four agent frameworks, with LangGraph suffering the worst impact. The work highlights a tradeoff where stronger guardrail reasoning increases resource use and introduces concentration risk for shared governance.

🛡️ In response to widespread prompt-injection risks, the article outlines runtime signals to detect compromised AI agents that possess the so-called lethal trifecta: access to private data, ingestion of untrusted content, and external communication ability. It argues that this trifecta is now the default for useful agents, so defenses must shift from architecture rules to behavioral, runtime detection. Recommended signals include instruction-following anomalies, unexpected tool-call sequences, low-bandwidth exfiltration channels, out-of-scope credential access, and suspicious memory writes.

📘 Today Google Cloud introduces the Open Knowledge Format (OKF), an open, vendor-neutral specification that formalizes the LLM-wiki pattern into a portable directory of markdown files with YAML frontmatter. OKF v0.1 defines a small set of conventions so different producers’ wikis can be consumed by agents without translation. The spec is intentionally minimal — one required type field per concept — and is accompanied by reference producer and consumer implementations and sample bundles.

🛡️ Cybersecurity researchers at Tenet Security disclosed a new attack class called Agentjacking that tricks AI coding agents into executing arbitrary code. The exploit leverages Sentry's public DSN and its MCP interaction to inject crafted error events, which agents like Claude Code and Cursor interpret as trusted resolution steps. Successful exploitation can expose sensitive data and run code with developers' privileges.

🛡️ Two security teams demonstrated attacks against OpenClaw, where hidden instructions in shared contacts, vCards, and location pins or ordinary-looking emails caused the agent to execute attacker-controlled code or exfiltrate sensitive data. Imperva found a message-object prompt-injection flaw that OpenClaw patched in version 2026.4.23, while Varonis showed social-engineering 'agent phishing' that requires architectural controls rather than a simple patch. Operators are urged to update, restrict outbound actions, and treat agents as junior employees needing human oversight.

🔎 AI agent skills can install third-party capabilities with privileged access, yet registries lack automated audits. Palo Alto Networks introduces Behavioral Integrity Verification (BIV), which compares declared metadata, executable code and natural-language instructions to detect mismatches. Applied to the OpenClaw registry, BIV found widespread deviations and identified multi-stage attack chains that enable credential theft, RCE and exfiltration. The report recommends inventorying skills and requiring pre-install behavioral checks.

🔍 AI agents can help SRE teams with incident response, triage and automation, but trust is granted only when agents demonstrate reliability under real-world stress. Teams need robust observability, explicit guardrails, human-in-the-loop workflows and explainability so recommendations are evidence-backed rather than speculative. Progressive autonomy, post-incident evaluation and compatibility with existing tools are essential for safe adoption.

🛡️ Researchers describe a new "agentjacking" attack that tricks AI coding agents into executing arbitrary code by injecting malicious instructions into Sentry error events. Tenet Security says the flaw leverages Sentry DSNs — public, write-only credentials — to post crafted markdown that appears as legitimate remediation guidance. Agents retrieving unresolved errors via MCP render the injected content as trusted and may execute the embedded commands with developer privileges. The report confirmed high exploitability across popular agents and thousands of exposed DSNs.

🧭 ASSERT is an open-source framework that converts natural-language behavior specifications into executable evaluation pipelines, generating test scenarios, datasets, metrics, and scorecards for models, agents, or applications. The pipeline systematizes intent into a concept spec, produces an editable behavior taxonomy, generates stratified test cases, records full inference traces, and scores each trace with rationales and policy citations. Internal validation showed ASSERT improves coverage, surfaces distinct failure patterns, and yields judge agreement with humans in most cases, while SME review confirmed alignment and credibility.

🛡️ An enterprise sales team built an AI agent to manage renewals; the agent reads emails, queries CRM data, drafts responses, and updates records. This workflow combines private data, untrusted input, and external communication, changing the security model. Traditional controls like IAM and DLP still matter but are insufficient alone. Runtime, context-aware controls that inspect prompts, outputs, and tool calls are required to prevent prompt injection, data exfiltration, and unsafe actions.

🔒 Varonis tested an OpenClaw-based AI agent named Pinchy with access to a controlled Google Workspace to see whether autonomous agents could be phished. The agent was given Gmail access plus mock AWS credentials, CRM exports, internal chats, and calendars, and it still leaked credentials and customer data in scenarios that mimicked routine colleague requests. A stricter safety profile improved performance, but the agent still failed when social trust cues were abused. Researchers say the problem stems from architecture and governance gaps, urging enforceable controls, identity segregation, and human review for sensitive requests.

📧 Researchers at Varonis tested an OpenClaw AI email agent connected to Gmail, browser tools, and internal data sources and found it vulnerable to common phishing techniques. The agent ran in both generic and strict configurations and used Google Gemini 3.1 Pro and OpenAI GPT-5.4 models. While the agent detected malicious links and OAuth apps, it still exfiltrated credentials and CRM data in scenarios exploiting identity verification failures. Varonis recommends explicit sender verification, restricted external emailing, and human approval for high-risk actions.

🤖 Microsoft has integrated Anthropic’s Claude Fable 5 into Foundry, bringing Mythos-level capabilities to GitHub Copilot and Foundry Agent Service with enterprise-grade safeguards. The model excels at long-running, multi-stage tasks—code refactors, deep research, and document-heavy workflows—while Foundry adds governance, observability, and deployment controls. Combined with Microsoft IQ, Fable 5 can reason across organizational data and applications to support production-grade autonomous agents.

🛡️ Google details how Google Security Operations pairs with Google AI Threat Defense to detect, investigate, and contain AI-accelerated attacks across cloud and enterprise environments. The post explains three specialized agents — Detection Engineering, Triage and Investigation, and Threat Hunting — that translate threat intelligence into custom detections, autonomously investigate alerts, and proactively hunt stealthy compromises. These agents use diverse telemetry, simulated events, and AI-driven automation to reduce time-to-detection and speed remediation, addressing gaps where patching is impossible or delayed.

🛡️ Researchers at the University of Toronto built an AI-driven worm prototype that autonomously discovered and exploited vulnerabilities across a simulated enterprise network. Using a locally hosted, free LLM and a custom agentic harness, the worm self-replicated to multiple systems by chaining old and recent CVEs and common misconfigurations. Over several days it spread to most targets, demonstrating that attackers do not need cutting-edge models to mount damaging, adaptive attacks. The findings underscore the need for faster patching, AI-assisted defensive testing, and improved architecture such as segmentation and zero trust.

🖥️ Amazon Bedrock AgentCore Runtime introduces the InvokeAgentRuntimeCommandShell API, providing a persistent, PTY-backed terminal over WebSocket into running agent sessions. This complements existing one-shot execution via InvokeAgentRuntimeCommand and delivers a full terminal experience inside an isolated microVM with features like colors, tab completion, Ctrl+C, resize, and automatic reconnect. Developers hosting coding agents (for example, Claude Code, OpenAI Codex, Amazon Kiro) can now authenticate, drop into the agent microVM, inspect files, run ad-hoc commands, and debug while retaining session state across reconnects. Each interactive session uses a runtime session ID and shell ID for resume; up to 10 concurrent shells are supported per runtime.

🔍 Microsoft has expanded its Taxonomy of Failure Modes in Agentic AI Systems with seven newly identified ways agentic AI can be compromised. The update cites rapid adoption, maturation of the Model Context Protocol (MCP) ecosystem, proliferation of computer-use agents, and increased empirical evidence as drivers. New failure modes include supply chain compromise, goal hijacking, inter-agent trust escalation, visual attacks on CUAs, session context contamination, MCP/plugin abuse, and capability disclosure. Microsoft recommends inventorying agent supply chains, issuing cryptographic attestations, adding these modes to red-team exercises, and auditing human-in-the-loop controls.