Wiz proposes a 4-pillar AI Threat Readiness Framework: (1) reduce attack surface and AI-scan all exposures; (2) accelerate CVE patching and zero-day response; (3) perform deep AI-driven code analysis; (4) automate real-time threat detection and containment.
https://www.wiz.io/blog/ai-threat-readiness-framework
#AI
Please open Telegram to view this post
VIEW IN TELEGRAM
❤1👍1🔥1
Doyensec compared Aikido Attack AI Pentest and XBOW Lightspeed for web app vulnerability detection, evaluating true/false positives, configuration, report quality, cost, speed, and impact on tested applications. Full findings available as a PDF.
#AI
Please open Telegram to view this post
VIEW IN TELEGRAM
❤1👍1🔥1
A guide to understanding Copilot Studio AI agents, their deeper architecture on Entra ID and APIM, and key security risks.
https://www.beyondtrust.com/blog/entry/copilot-studio-ai-agents-security-risks
#AI
Please open Telegram to view this post
VIEW IN TELEGRAM
❤1👍1🔥1
Any web page a victim asks ChatGPT to summarize can become a phishing payload. P0 Labs research reveals a Markdown rendering vulnerability in ChatGPT's response UI.
https://permiso.io/blog/chatgpt-markdown-rendering-vulnerability
#AI
Please open Telegram to view this post
VIEW IN TELEGRAM
❤1👍1🔥1
Run a fully self-contained, self-improving AI agent from a single folder or USB drive.
https://github.com/techjarves/hermes-usb-portable
#AI
Please open Telegram to view this post
VIEW IN TELEGRAM
👍2❤1🔥1
Entra Agent ID is an extension of Entra's application model that provides identities for AI agents. Unlike applications, the agent identity model allows linking a single app registration (blueprint) to multiple identities and their associated privileges, increasing the potential blast radius of a compromised agent.
https://securitylabs.datadoghq.com/articles/agent-id-blueprint-blast-radius
#AI
Please open Telegram to view this post
VIEW IN TELEGRAM
❤1👍1🔥1
Mitiga Labs scanned 50,000+ AI instruction files across 7,000+ repos, finding prompt-exfiltration malware, ANTHROPIC_BASE_URL MITM overrides, YOLO-mode permission bypasses, and 1,230+ hardcoded API keys. They released Skillgate, a free scanner, to detect these techniques.
https://www.mitiga.io/blog/malware-in-ai-instruction-files-skillgate
#AI
Please open Telegram to view this post
VIEW IN TELEGRAM
❤1👍1🔥1
NCC Group whitepaper covering security of AI coding agents (Claude Code, Cursor, Codex) as of May 2026. Analyzes permission models, OS-level sandboxing, agent tool attack surfaces, and common vuln classes: workspace trust bypasses, sandbox escapes, permission prompt bypasses, sensitive file overwrites, and indirect prompt injection.
#AI
Please open Telegram to view this post
VIEW IN TELEGRAM
❤2👍2🔥2
Sysdig TRT caught a threat actor using a stolen Ollama server to power an autonomous, multi-stage offensive hacking tool.
https://www.sysdig.com/blog/llmjacking-evolved-attackers-are-using-stolen-ai-compute-to-build-offensive-agentic-tools
#AI
Please open Telegram to view this post
VIEW IN TELEGRAM
❤2👍1🔥1
Unlike traditional SAST tools that rely solely on parsing and analysis rules, this project uses LLM (e.g. Claude from Anthropic, GPT from OpenAI or Gemini from Google) to find vulnerabilities.
https://github.com/DataDog/datadog-saist
#AI
Please open Telegram to view this post
VIEW IN TELEGRAM
❤1👍1🔥1
Threat actors impersonate AI brands (ChatGPT, Claude, DeepSeek, Copilot) in phishing, malvertising, and SEO-poisoning campaigns to steal credentials, credit card data, and deliver infostealers like Vidar. Storm-3075 and Fox Tempest are attributed actors using signed malware.
https://www.microsoft.com/en-us/security/blog/2026/06/08/ai-brands-as-bait-how-threat-actors-are-using-the-ai-hype-in-social-engineering
#AI
Please open Telegram to view this post
VIEW IN TELEGRAM
❤1👍1🔥1
This post compares three AI agent identity models: acting as the user (simple but single-player), service account tokens (common in production but insecure), and SPIFFE-based workload identity (best but costly to implement). Covers governance plumbing like Okta XAA and cloud-provider managed options.
https://kanenarraway.com/posts/agent-identity-models
#AI
Please open Telegram to view this post
VIEW IN TELEGRAM
Please open Telegram to view this post
VIEW IN TELEGRAM
❤2👍1🔥1
Uber extended its Zero Trust architecture to assign SPIRE-backed cryptographic identities to AI agents, issuing short-lived, single-hop JWTs via a Security Token Service. Full actor chains (user to agent to tool) propagate across multi-agent MCP calls, enabling end-to-end auditability with sub-40ms P99 token exchange latency.
https://www.uber.com/us/en/blog/solving-the-agent-identity-crisis
#AI
Please open Telegram to view this post
VIEW IN TELEGRAM
❤1👍1🔥1
Anthropic eBook applying Zero Trust principles to enterprise autonomous AI agents, covering threat taxonomy (prompt injection, tool/MCP hijacking, memory poisoning, supply chain risks), tiered controls (Foundation/Enterprise/Advanced) for identity, least-privilege, observability, and AI-accelerated defensive operations.
#AI
Please open Telegram to view this post
VIEW IN TELEGRAM
❤1👍1🔥1
Noma Labs found a prompt injection flaw (GitLost) in GitHub Agentic Workflows: posting a crafted issue in a public org repo causes the AI agent to exfiltrate private repository contents as a public comment, bypassing guardrails via the keyword "Additionally.".
https://noma.security/blog/gitlost-how-we-tricked-githubs-ai-agent-into-leaking-private-repos
#AI
Please open Telegram to view this post
VIEW IN TELEGRAM
❤1👍1🔥1
GhostApproval is a symlink-following vulnerability affecting 6 AI coding assistants (Amazon Q, Claude Code, Augment, Cursor, Google Antigravity, Windsurf). Malicious repos trick agents into writing outside the workspace sandbox, enabling RCE. Critically, agents internally recognize the dangerous target but confirmation prompts hide it from users, nullifying Human-in-the-Loop protections.
https://www.wiz.io/blog/ghostapproval-a-trust-boundary-gap-in-ai-coding-assistants
#AI
Please open Telegram to view this post
VIEW IN TELEGRAM
Please open Telegram to view this post
VIEW IN TELEGRAM
🔥3❤1👍1
Anthropic's Deputy CISO shares a four-question framework for assessing agentic AI risk, and walks through controls that keep agent deployments bounded and auditable.
https://claude.com/blog/ciso-guide-to-agentic-ai
#AI
Please open Telegram to view this post
VIEW IN TELEGRAM
❤1👍1🔥1
A reference architecture for giving a locally-running AI agent a trustworthy, auditable identity, without long-lived credentials on disk, including a structural defense against prompt injection built into the protocol layer.
https://1password.com/blog/ai-agent-identity-delegated-local
#AI
Please open Telegram to view this post
VIEW IN TELEGRAM
Please open Telegram to view this post
VIEW IN TELEGRAM
🔥2❤1👍1
A study of ~200,000 LLM responses found 5 frontier models (Claude, GPT, Gemini, DeepSeek) hallucinate nonexistent package names at 4.62-6.10%, with 53 shared fictitious names on PyPI/npm still registrable and exploitable via slopsquatting attacks.
https://socket.dev/blog/slopsquatting-targets-across-frontier-llms
#AI
Please open Telegram to view this post
VIEW IN TELEGRAM
❤1👍1🔥1
Permiso researchers deployed an AI agent (Rufio) into the OpenClaw ecosystem and found active malware campaigns in its unvetted skill marketplace (ClawHub), credential-harvesting skills with 377+ downloads, C2 infrastructure, and prompt injection attacks targeting agents holding plaintext credentials to email, Slack, and file systems.
https://permiso.io/blog/inside-the-openclaw-ecosystem-ai-agents-with-privileged-credentials
#AI
Please open Telegram to view this post
VIEW IN TELEGRAM
❤1👍1🔥1