All news with #ai agent hijacking tag

🛡️ 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.

🔍 New research finds current AI web agents largely fail to defend against prompt injection attacks. The StakeBench benchmark tested GPT‑5 and Gemini‑powered agents across realistic web scenarios, revealing high success rates for both direct and indirect injections and exposing failure modes like stealthy parasitism and misaligned disruption. Results show vulnerabilities vary by stakeholder and agent architecture.

🔎 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.

🤖 AWS Step Functions now integrates with the Amazon Bedrock AgentCore managed harness (preview) to add AI agent reasoning steps to workflows. The integration lets you declare agents via configuration, run agents in parallel or sequence, add human approvals, and view execution history with agent inputs, outputs, token usage, and CloudWatch links. You can reuse or create harnesses from Workflow Studio, apply per-invocation overrides, and persist agent context with session IDs. The harness preview and integration are available in select regions and standard Step Functions and Bedrock pricing applies.

🔒 A wave of account takeovers targeted high-profile Instagram profiles after attackers shared instructions for tricking Meta’s AI support assistant into relinking accounts to attacker-controlled email addresses. The technique, circulated on Telegram on May 31, reportedly involved using a VPN to appear from the target’s locale, initiating a password reset, and persuading the AI bot to add a new email. Meta acknowledged a brief compromise of a dormant Obama White House account and pushed an emergency patch while asserting no backend database was breached. Experts warn AI-driven support flows introduce new attack surface and recommend strong MFA such as passkeys or security keys to mitigate risk.

🛰️ The Linux Foundation has proposed DNS-AID, a standards-driven extension to the Domain Name System to let AI agents discover, verify, and communicate without new infrastructure. The project leverages a well-known DNS address pattern (for example, _index._agents.{domain}) to provide a global, vendor-neutral directory for agents and MCP servers. Initial work was done by Infoblox, with contributions from Deutsche Telekom and Amazon, and the foundation is soliciting further input to keep the approach scalable and secure.

🛡️ Microsoft uncovered a targeted cryptojacking campaign that lures owners of high-performance PCs to malicious download pages for utilities like CrystalDiskInfo and HWMonitor. The attackers used SEO poisoning and, in some cases, manipulated AI chatbots to surface attacker-controlled download links. Infected ZIP archives include legitimate utilities and a malicious DLL that installs the ScreenConnect remote access tool, enabling persistent access and deployment of a process-hollowing loader that ultimately launches GPU miners.

🔒 PraisonAI contained a critical authentication bypass (CVE-2026-44338) in its legacy Flask API server that sets AUTH_ENABLED = False and AUTH_TOKEN = None by default. Exploitation allows unauthenticated callers to enumerate configured agents via /agents and to trigger workflows through /chat, potentially consuming model quotas and exposing run results. The flaw affects versions 2.5.6–4.6.33 and was fixed in v4.6.34; operators are advised to update, audit deployments, and rotate exposed credentials.

🔒 Knostic’s internet-wide reconnaissance discovered 1,862 exposed MCP servers, and manual checks of 119 instances showed every sampled server returned internal tool listings without authentication. High-impact flaws like EchoLeak (CVE-2025-32711) and mcp-remote (CVE-2025-6514) illustrate how poisoned documents and command-injection in widely used packages can enable silent data exfiltration or full system compromise. The article prescribes immediate adoption of zero-trust controls: authentication on every interaction, network segmentation, cryptographic signing for tool definitions, continuous integrity monitoring, and human approval for sensitive actions.

🛡️ Analysts and Orchid Security warn that enterprises are deploying AI agents faster than governance can keep up, creating an invisible layer of "identity dark matter" that conventional IAM misses. Orchid Security inspects applications at the binary and configuration layer to discover agents, audit compliance, and locate static credentials. Its Ask Orchid assistant answers natural-language questions about active agents, NIST compliance, and credential risks, then recommends prioritized remediation. This in-application observability aims to close the structural gap in identity visibility and enforce purpose-bound, least-privilege controls.

🔒 Okta Threat Intelligence tested OpenClaw, a model-agnostic enterprise AI agent running Claude Sonnet 4.6, and found it could be manipulated to disclose sensitive credentials. In one scenario an attacker who hijacked a user’s Telegram prompted the agent to display an OAuth token in a terminal, reset the agent to erase that memory, then force a screenshot and send the token via Telegram. Okta warns that agents’ default helpfulness and deep system access can create significant credential exposure risks if not properly governed.

⚠️ Security researchers disclosed a high-severity vulnerability in the Cursor AI-powered IDE that can lead to arbitrary code execution when its agent interacts with a malicious repository. Novee Security's analysis shows an attacker can embed a bare Git repository with a crafted hook and trigger it when the IDE autonomously runs Git operations. Cursor patched the flaw in version 2.5; there are no reports of active exploitation.

🛡️At VirusTotal we are integrating reputation and Code Insight directly into AI agent decision loops so agents can consult verdicts and context as part of their runtime behavior. Two community plugins, VT-sentinel (OpenClaw) and hermes-virustotal (Hermes), demonstrate the approach using the new VTAI API with compact responses and per-agent identities. Both MIT-licensed projects scan files, annotate hashes, and provide configurable privacy and enforcement presets so agents can quarantine, block, or proceed based on risk appetite.

🔒 Multiple vendors and researchers this week disclosed a broad set of active threats spanning cloud environments, endpoints, and messaging platforms. OAuth consent abuse campaigns impersonated trusted apps to harvest tokens and access mail and files without passwords, while the BlackSanta campaign used resume-themed ISOs to chain DLL side‑loading and disable AV/EDR via vulnerable drivers. Other notable items include microcontroller debug bypasses, ZIP header evasion that defeats some AV/EDR tools, an AI-agent compromise of an internal platform, and targeted phishing against Signal and WhatsApp users.

🛡️ Attackers increasingly treat high-volume phishing as a weapon, flooding Security Operations Centers to exhaust analysts and hide targeted spear-phish. The article argues defenders must move from rule-based automation to decision-ready investigations—transparent, auditable agentic AI that produces concise verdicts and evidence. This reduces analyst fatigue, restores rapid response, and limits the window for attacker success.

🛡️ CodeWall’s autonomous red-team agent compromised hiring startup Jack & Jill by chaining four seemingly minor bugs into a complete account takeover within an hour. The agent abused a permissive URL fetcher, an enabled test-login mode, missing onboarding role checks, and absent domain verification to map APIs, authenticate via a test OTP flow, and escalate to org-admin privileges. It then generated synthetic voice clips to social-engineer Jack, conducting 28 multi-turn exchanges and even impersonating Donald Trump before moving on, demonstrating how AI can rapidly combine low-risk flaws into high-impact attacks.

🔒 OpenAI said it will acquire AI testing startup Promptfoo to strengthen security checks for AI agents as enterprises deploy autonomous systems in business workflows. Promptfoo’s tools let developers test LLM applications against adversarial prompts, including prompt injection and jailbreak attempts, and evaluate whether models follow safety and reliability guidelines. OpenAI plans to integrate Promptfoo into OpenAI Frontier and to continue developing the open-source project while expanding enterprise capabilities.

🤖 AI-based assistants such as OpenClaw are rapidly reshaping organizational security, blurring boundaries between data and code and between trusted co-workers and insider threats. Incidents and research show agents taking autonomous actions and misconfigured admin interfaces exposing credentials, conversations, and integrations. Demonstrated supply-chain and prompt injection attacks can install rogue agents and manipulate agent perception. Organizations should isolate agents, enforce strict network controls, vet third-party skills, and address AI fragility as a core security concern.

🚀 Google Cloud and Nokia announced an integration at MWC Barcelona that connects Nokia Network as Code (NaC) with Google Cloud’s agentic AI stack to enable AI agents to observe, program, and optimize mobile networks autonomously. The collaboration leverages Gemini models and standardized protocols such as A2A and MCP to translate natural-language intent into network actions. An Agent Development Kit (ADK) allows enterprises to build custom multi-agent workflows that bridge business logic and network intelligence, delivering a zero-code, intent-driven developer experience.

🔍 OpenClaw is an AI-driven automation framework with a modular skills marketplace that lets agents run user-installed plugins to manage mail, schedules, and system tasks. Security researchers disclosed multiple critical flaws — including one-click RCE (CVE-2026-25253), token/OAuth abuse, prompt-injection pathways, and absent sandboxing — and documented dozens of poisoned skills on ClawHub. Flare's telemetry shows significant chatter across research and fringe channels but limited evidence of mass criminal operationalization; the immediate confirmed threat is supply-chain abuse where malicious skills execute with agent-level privileges and exfiltrate credentials and sessions.