Zenity

🚨 Agentic Browsers don’t just read the web. They reason, decide, and take action with full user authority. That’s exactly what turns a harmless browsing session into a high-impact breach. 🚨 Zenity Labs found that a single malicious instruction hidden inside everyday content, like a Slack thread, a Google Doc, even a product review, can trigger actions the user never intended. 🤯 File access. SaaS pivots. Cloud credential use. Immediate exfiltration. 🕵️ Enterprise compromise at the speed of an agent. 👀 Get the full breakdown and see where your exposure sits. 👉 https://lnkd.in/efvcffG5 #AIAgentSecurity #AgenticSecurity #AIThreats"

Financial services are not just experimenting with AI agents anymore, they are wiring them into core workflows and customer journeys. ⚙️ That shift raises tough questions for security, compliance, and risk teams who have to keep Microsoft 365 Copilot, Copilot Studio, and Azure AI Foundry deployments safe, governed, and auditable. 🧩 On December 10, Kayla Underkoffler, Ryan R. (Slalom), and Zohar Raz (Microsoft) will share how leading financial institutions are building AI agents with defense in depth, clear guardrails, and real visibility into behavior across the Microsoft ecosystem. 🏦 🔗 https://lnkd.in/eJya4yBG #AIAgentSecurity #MicrosoftCopilot

The White House’s new Genesis Mission is a major milestone — a national push to accelerate scientific breakthroughs using AI. 🚀 But with unprecedented access to sensitive research data and autonomous AI agents, it also creates a new security imperative. 🔒 In our latest post, we break down why large-scale, agent-driven AI initiatives like Genesis need secure-by-design guardrails, not retrofitted legacy controls — and what that looks like in practice. 🛡️ Read the full breakdown 👉 https://lnkd.in/gVTEea9d #AI #Cybersecurity #AIAgents #AIsecurity #TechPolicy #NationalSecurity #Zenity

After a full day of Black Hat Europe briefings, the real conversations happen offstage. 👥 On December 10th, Zenity is teaming up with Tines, Cyera, Semgrep, and Sublime Security to host a Happy Hour aboard the Sunborn superyacht, right next to ExCeL London. 🧠🍸 If you’re in town for Black Hat, this is your chance to connect with security leaders, swap insights on agentic risk, and unwind with the teams building the next era of enterprise security. 🤩 Reserve your spot 👉 https://lnkd.in/eGdtZ8uF #BlackHatEurope #AIAgentSecurity #AISecurity #AgenticSecurity

Claude Code wasn’t designed to attack anyone. But Anthropic’s latest disclosure shows how easily it can be weaponized when put in the wrong hands. 🚨 GTG 1002 tricked Claude into believing it was a legitimate penetration tester, enabling it to automate reconnaissance, exploitation, credential harvesting, and exfiltration across more than 30 organizations. This was an AI agent acting as the attacker, not just assisting one. ⚠️ The message is clear. Any advanced coding agent inside your environment can become a threat. Claude inherits employee permissions, interacts with repos and CI pipelines, executes shell commands, and integrates with MCP servers that can quietly expand its reach. The blast radius is massive. 🔍 Zenity provides the visibility and control needed to detect and stop a rogue coding agent acting from inside your organization. The next GTG 1002 won’t come through your perimeter. It will run on a trusted machine, using trusted tools, with trusted credentials. 🧩 Read the full analysis and steps to take now 👉 https://lnkd.in/ep3JcffY #AIAgentSecurity #AgenticSecurity #AISecurity #ClaudeCode #GTG1002 #Cybersecurity

⚠️ Agentic Browsers collapse the boundary between “local incident” and “enterprise breach.” Once hijacked, they move across SaaS, cloud, and endpoints as the user. ⚠️ Zenity Labs shows how a single malicious instruction hiding in a calendar invite, Slack thread, or website can trigger: • AWS access • Filesystem scraping • Full data exfiltration • Cross-system pivoting No exploit. Just trust. And that’s the problem. 🔥 See the full attack paths in the report and benchmark your exposure with a risk assessment. 👁️ 👉 https://lnkd.in/e_mKhFHc #AIAgentSecurity #AgenticSecurity #AIsecurity #AIThreats #BHEMEA

Today’s agent defenses trace tool calls, monitor I/O, and limit rates, but the LLM itself remains a black box. 🧠 Zenity Labs is exploring how to open that box. Using attribution graphs, we can visualize how jailbreaks and prompt injections form inside the model — showing interpretable features like “regret,” “illegal,” “assistant,” and “cyber” activating as the model decides whether to comply or refuse. 🔍 It’s an early step toward understanding how internal model circuits shape security behavior and how we might one day monitor or steer them directly. ⚙️ Full research post 👉 https://lnkd.in/gWTGKX-Z #ZenityLabs #AIAgentSecurity #MechanisticInterpretability #PromptInjection #Jailbreaks #AIResearch #AgenticSecurity

Michael Bargury sat down with Calcalist כלכליסט for a wide-ranging conversation on the rise of agentic AI, the evolving threat landscape, and what it takes to build a cybersecurity company in an era where AI systems act, decide, and move at machine speed. 🎙️ He shares his path from discovering cyber while playing Counter-Strike, to Unit 8200, to six years at Microsoft, to founding Zenity — and why securing AI agents is quickly becoming one of the most urgent challenges facing global enterprises. 🔍 They also dig into recent Zenity Labs research, revealing how vulnerabilities in OpenAI systems could allow impersonation, data leakage, or even malware delivery through ChatGPT-style interfaces. 🚀 Give it a listen here: 🎧 Spotify: https://lnkd.in/gFXjNXFS 🎧 Calcalist: https://lnkd.in/gwhBcr6C #AIAgentSecurity #Cybersecurity #AI #IsraelTech #Zenity

What your agentic platform says about you.... The race to adopt AI in financial services is accelerating, but with so many options, how do you choose the right platform for your organization? In our latest blog, to illustrate these choices, we break down how Microsoft Copilot Studio and Foundry compare, including: ✅ The platforms best suited to build particular agents ✅ Governance and compliance requirements ✅ Integration with existing ecosystem tools ✅ Security implications for sensitive financial data ✅ Scalability for enterprise AI adoption If you’re evaluating AI solutions, this blog will shed some light on how to balance innovation with risk management. 👉 Read the full article here: https://lnkd.in/edqvVMQT #AIAgentSecurity #AIAgent #CopilotStudio #Foundry