AppOmni

You can't contain AI with prompts alone. Whether it's McDonald's customer service bots being jailbroken to bypass their purpose (to debug Python code) or Now Assist AI agents being tricked into recruiting more powerful agents (research link in comments), the vulnerability is the same: prompt injection. https://hubs.ly/Q04f7NMH0 LLMs understand fluent language, not rigid rules. Hidden guardrails are just suggestions to determined attackers. In multi-agent systems, it's exponentially worse. One compromised agent can recruit others and expand the blast radius. The fix isn't better prompting; it's governance: supervised execution, configuration controls, and real-time monitoring. If you're deploying AI at scale, treat prompt injection like you would any other security vulnerability.

I'm excited to be speaking in front of my Philly peeps next week at SecureWorld!! The panel session is around leveraging AI in Cyber Defense. We'll be tackling the topic of AI being a very powerful tool that must be leveraged in cyber defense, but in parallel, the same technology is being leveraged by attackers to move at speeds we've never seen and devise new attack paths not previously possible....a double edged sword. If you're attending SecureWorld come join the conversation on Wed (5/6) from 1:10-1:45pm!

AI is changing how SaaS environments are attacked, right now. Take the EvilToken campaign for example. Attackers used AI-driven lures to hijack real Microsoft login processes, tricking users into granting valid OAuth tokens. ‼️ 𝑾𝒉𝒚 𝒕𝒉𝒊𝒔 𝒎𝒂𝒕𝒕𝒆𝒓𝒔: attackers are using AI to accelerate their attacks, increase their effectiveness, and even exploit trust between systems. This isn’t the only way cyber criminals are weaponizing AI. We break down the 3 most common AI-driven attack paths in SaaS and how to defend against them with identity-first, continuous security. https://hubs.la/Q04d_X160

Security leaders: We want to hear from you. 👂

AI in cybersecurity is only as good as the data behind it. ☝️ AppOmni Director of AI, Melissa R., explains why domain expertise matters. 📽️ Watch to see why you can’t rely on AI models to “figure out” bad or noisy data on their own. SOON: On our May 27 webinar, Melissa shares more about responsible AI in cybersecurity - link in comments to register.

💙 This #NationalVolunteerMonth, we're featuring Vivek Kumar, our Sr. Director of Cloud & Tech Alliances, and his work with the EmpowerKids Foundation. Vivek founded the organization to provide quality education to children in rural India. From building a 5-room elementary school to running a coaching center for competitive exams, Vivek’s mission is simple: break the cycle of poverty through learning. 📚 To date, the foundation has helped nearly 500 children, including several girls now on their way to higher education. We’re honored to have leaders like Vivek who drive impact far beyond the tech world.

Unique opportunity for a highly experienced Data Scientist to define and build ML-driven risk scoring, prioritization, and AI-powered security workflows within AppOmni’s SaaS security platform. Must have hands-on ML experience in security and a passion for applying AI to real-world threats 🚀 Remote-USA #DataScience #ArtificialIntelligence #Risk #Security #SaaS #SSPM #AI Find out more below; https://lnkd.in/g4DNK9zw

🗣️ "No one can ever keep anything 100% out of attackers' hands. The best that can be done is to make it more difficult for them to get access to it." — Melissa R., Director of AI at AppOmni in Dark Reading. https://hubs.la/Q04c3z8b0 As Anthropic introduces initiatives like Project Glasswing to put powerful capabilities in defenders’ hands, this remains true: AI can be a powerful force for both defenders and attackers. Efforts like Glasswing are a meaningful step forward, helping security teams stay ahead as these capabilities evolve. While it’s impossible to keep attackers out completely, the advantage goes to teams that: 📉 Reduce access 🙅 Enforce least privilege 🔍 Monitor continuously What are your thoughts on the evolution of AI security? Let us know in the comments.