AI-Powered Cyberattacks: Exaforce's Automated Security Solution

If your AI agent can log in, pull files, and take action… are you protecting it like software, or like a privileged employee? Because that difference is where the new risk lives. In 2026, these agents aren’t “chatbots with a nicer UI.”  They authenticate, call APIs, touch databases, and execute business logic. And Gartner reported about 𝟔𝟎%+ 𝐨𝐟 𝐥𝐚𝐫𝐠𝐞 𝐞𝐧𝐭𝐞𝐫𝐩𝐫𝐢𝐬𝐞𝐬 𝐧𝐨𝐰 𝐫𝐮𝐧 𝐚𝐮𝐭𝐨𝐧𝐨𝐦𝐨𝐮𝐬 𝐚𝐠𝐞𝐧𝐭𝐬 𝐢𝐧 𝐩𝐫𝐨𝐝𝐮𝐜𝐭𝐢𝐨𝐧, up from 𝟏𝟓% 𝐢𝐧 𝟐𝟎𝟐𝟑. So what’s the backlash really about? It’s not “AI is risky” (we knew that). It’s that 𝐚𝐠𝐞𝐧𝐭𝐢𝐜 𝐀𝐈 𝐜𝐫𝐞𝐚𝐭𝐞𝐬 𝐫𝐢𝐬𝐤𝐬 𝐭𝐫𝐚𝐝𝐢𝐭𝐢𝐨𝐧𝐚𝐥 𝐬𝐞𝐜𝐮𝐫𝐢𝐭𝐲 𝐰𝐚𝐬𝐧’𝐭 𝐝𝐞𝐬𝐢𝐠𝐧𝐞𝐝 𝐭𝐨 𝐬𝐞𝐞: 1. 𝐓𝐡𝐞 𝐬𝐞𝐦𝐚𝐧𝐭𝐢𝐜 𝐚𝐭𝐭𝐚𝐜𝐤 𝐬𝐮𝐫𝐟𝐚𝐜𝐞: prompt injection doesn’t break a firewall. It persuades the agent to ignore rules.  2. 𝐈𝐝𝐞𝐧𝐭𝐢𝐭𝐲 𝐛𝐞𝐜𝐨𝐦𝐞𝐬 𝐭𝐡𝐞 𝐛𝐫𝐞𝐚𝐜𝐡: spoof an agent identity or steal a token, and you don’t just get access, you get automated execution.  3. 𝐀𝐏𝐈 𝐜𝐡𝐚𝐢𝐧𝐢𝐧𝐠 𝐚𝐦𝐩𝐥𝐢𝐟𝐢𝐞𝐬 𝐛𝐥𝐚𝐬𝐭 𝐫𝐚𝐝𝐢𝐮𝐬: one agent action can trigger ten downstream actions across SaaS. And considering its business impact,  this isn’t “one compromised user.”  It’s 𝐚 𝐜𝐨𝐦𝐩𝐫𝐨𝐦𝐢𝐬𝐞𝐝 𝐨𝐩𝐞𝐫𝐚𝐭𝐨𝐫 𝐭𝐡𝐚𝐭 𝐜𝐚𝐧 𝐦𝐨𝐯𝐞 𝐦𝐨𝐧𝐞𝐲, 𝐝𝐚𝐭𝐚, 𝐚𝐧𝐝 𝐚𝐩𝐩𝐫𝐨𝐯𝐚𝐥𝐬 𝐚𝐭 𝐦𝐚𝐜𝐡𝐢𝐧𝐞 𝐬𝐩𝐞𝐞𝐝. So, now on,  1. 𝐆𝐢𝐯𝐞 𝐞𝐯𝐞𝐫𝐲 𝐚𝐠𝐞𝐧𝐭 𝐚 𝐫𝐞𝐚𝐥 𝐢𝐝𝐞𝐧𝐭𝐢𝐭𝐲 + 𝐥𝐞𝐚𝐬𝐭 𝐩𝐫𝐢𝐯𝐢𝐥𝐞𝐠𝐞 (separate accounts, short-lived tokens, rapid revocation).  2. 𝐌𝐨𝐧𝐢𝐭𝐨𝐫 𝐚𝐠𝐞𝐧𝐭 𝐛𝐞𝐡𝐚𝐯𝐢𝐨𝐫 𝐥𝐢𝐤𝐞 𝐲𝐨𝐮 𝐦𝐨𝐧𝐢𝐭𝐨𝐫 𝐢𝐧𝐬𝐢𝐝𝐞𝐫𝐬 (anomaly detection on access patterns + actions).  3. 𝐀𝐬𝐬𝐮𝐦𝐞 𝐩𝐫𝐨𝐦𝐩𝐭 𝐢𝐧𝐣𝐞𝐜𝐭𝐢𝐨𝐧 𝐞𝐱𝐢𝐬𝐭𝐬 and build layered defenses (input/output controls, guardrails, red-teaming). Because once an AI agent can act on behalf of your organization, 𝐬𝐞𝐜𝐮𝐫𝐢𝐭𝐲 𝐬𝐭𝐨𝐩𝐬 𝐛𝐞𝐢𝐧𝐠 𝐚𝐛𝐨𝐮𝐭 𝐩𝐫𝐨𝐭𝐞𝐜𝐭𝐢𝐨𝐧 𝐚𝐧𝐝 𝐬𝐭𝐚𝐫𝐭𝐬 𝐛𝐞𝐢𝐧𝐠 𝐚𝐛𝐨𝐮𝐭 𝐜𝐨𝐧𝐭𝐫𝐨𝐥. #AISecurity #CyberSecurity #AIGovernance #EnterpriseAI #RiskManagement

AI is changing how security teams work by automating alert triage and investigation. This allows analysts to focus on real threats and reduces the time it takes to respond to incidents. Security teams should consider AI solutions that offer depth, accuracy, transparency, adaptability, and workflow integration to improve their security operations. 🛡️ #CyberNewsLive https://lnkd.in/gwSDAEaB

You don’t need prompt injection to compromise agentic systems. While the security for AI industry has long focused on stopping prompt injection, a greater threat lurks beneath the surface. We’ve been trying to deflect individual bullets with tiny shields rather than building a concrete bunker. Last week, two findings surfaced that highlighted weaknesses in the architecture built around LLMs. Here’s an overview of the research to illustrate the point: 𝗗𝗼𝘂𝗯𝗹𝗲 𝗔𝗴𝗲𝗻𝘁: XM Cyber posted research on Google’s Vertex AI, “where default configurations allow low-privileged users to pivot into higher-privileged Service Agent roles.” It requires existing access to the environment, but shows a clear path to escalate privileges, a key component to many attacks. 𝗕𝗼𝗱𝘆𝗦𝗻𝗮𝘁𝗰𝗵𝗲𝗿: AppOmni pulled off some impressive research that combined multiple design flaws that would allow an unauthenticated attacker to impersonate any ServiceNow user and execute AI workflows, such as creating new users or exfiltrating data. When we know LLMs are inherently vulnerable to prompt injection and that there is no reliable way to prevent it, it becomes even more important to strengthen the surrounding systems. For defenders, ask yourself these questions: 1. Do you know what agents are running in your environment? 2. Do you know what tools and data those agents have access to? Have you threat modeled what could happen if someone malicious were using those agents? 3. Do you have the right visibility into what is happening on the systems running the agents, as well as with the agents themselves? Can you detect when agents are doing something suspicious? If you’re trying to answer these questions and wondering how, let’s talk. We’re tackling these emerging problems at Evoke Security.

What many CISOs are experiencing right now is acceleration, not reinvention. AI is sharpening familiar tactics: phishing that’s harder to detect, social engineering that’s more convincing, and deepfake capabilities that are increasingly accessible. On the cyberdefense side, the benefits are real. AI is helping teams reduce noise, spot patterns faster, and compress investigation time. Where the conversation gets harder, and more important, is governance. As agentic AI becomes embedded in enterprise environments, questions of ownership and accountability move to the forefront. Who owns risk when a synthetic identity initiates an attack? How do we protect LLMs that now concentrate sensitive data? What does “human in the loop” actually mean when decisions happen at machine speed? Frameworks are necessary, but they aren’t sufficient. What CISOs are asking for is clearer ownership, stronger threat modeling, measurable operational impact, and governance that boards can trust, not just vendor assurances. AI will continue to augment security teams, not replace them. But leadership judgment, accountability, and trust remain firmly human responsibilities. That’s the shift this article captures, and the one boards need to engage with now.

Attackers are using AI to study your business. 🧠 One-size-fits-all defenses will not survive mass-personalized cyberattacks. Preemptive security is becoming the new baseline. 👉 Learn what every security leader should prepare for: https://hubs.la/Q03ZxdvD0

A lot of people are wondering what AI will mean for cybersecurity jobs. 💡Our view: it won’t replace analysts—it’ll reshape the role. Doron Davidson sees two shifts. Analysts will become: 1️⃣ Customer-facing: translate agent-led analysis into clear, actionable guidance 2️⃣ Agent builders: identify tasks that AI agents can own and help develop them Hear more in his discussion on the N2K | CyberWire. Link in the comments 👇 #AgenticAI #SecOps

Conventional cybersecurity won’t protect your AI. As GenAI moves into core business operations, many leaders assume existing security controls will scale with it. Our research shows they won’t. Traditional defenses were built for deterministic software. GenAI is probabilistic, data-driven, and dependent on complex supply chains—creating new risk surfaces most organizations are not prepared for. The takeaway: securing AI isn’t about patching applications. It requires hardening the infrastructure and supply chains AI depends on—and increasingly, using AI itself as part of the defense. Now live in Harvard Business Review. Thanks to Juan Martinez and the HBR editorial team for their guidance, and to my colleague Ijlal Loutfi for contributing the case study. #aisecurity

Palo Alto Networks just classified AI agents as "the new insider threat." Their Chief Security Intelligence Officer dropped this stat: 60% of organizations can't shut down a rogue AI agent if they need to. We're at a weird inflection point. By end of 2026, 40% of enterprise apps will have AI agents integrated. That's 8x growth in one year. But most security teams are treating them like normal software when they're actually identities that need credentials, permissions, and access to your systems. A fintech company deployed an AI agent to automate compliance checks. Within 3 days it had accumulated read access to their production database, their S3 buckets, and their internal admin tools. Nobody planned that - it just requested permissions as it needed them, and the team kept approving because it was making their lives easier. Here's what worries me about the current state: - Government agencies: 90% lack purpose binding for AI systems, 76% have no kill-switch capabilities - Healthcare orgs: 77% haven't tested recovery objectives, 64% lack AI anomaly detection - Enterprise overall: Only 6% have an advanced AI security strategy This isn't theoretical anymore. Anthropic disrupted the first documented AI-orchestrated cyber espionage campaign in November 2025. We're seeing goal hijacking, privilege escalation, and autonomous lateral movement happening at machine speed - faster than humans can intervene. What makes this different from traditional insider threats is scale and speed. A compromised human might exfiltrate data over weeks. A compromised AI agent can make thousands of decisions per minute across your entire infrastructure. I don't have all the answers here, but I think we need to treat AI agents more like employees than tools. That means identity governance, kill-switches, intent auditing, and AI-specific anomaly detection. You wouldn't give a new hire unlimited system access on day one - why do that with an AI agent? Curious what others are seeing. Are your security teams treating AI agents differently than regular applications? What controls have you put in place? #AISecurity #CyberSecurity #EngineeringLeadership #InfoSec #AIAgents

No one logs in for the day thinking, “Today’s the day I cause an email breach.” And yet, that’s how most security incidents begin. Modern phishing and email compromise attacks aren’t obvious like they used to be. They come through as: 😧 Trusted vendors. 😕 Executives. 😬 Routine accounting requests. 🫣 Normal business conversations that slip through when teams are busy and moving fast. One click is all it takes, and once it happens, the financial and reputational impact for an organization escalates quickly. This week, we’re spotlighting Abnormal AI as a featured ONST partner. They bring a cutting-edge approach to email security, providing a critical layer of protection between organizations and attackers who rely on trust, context, and human behavior to succeed. At ONST Technologies, we’ve seen time and again that proactive security is far less expensive, and far less disruptive, than responding after the damage is done. Email remains the #1 attack vector across industries, and it isn’t slowing down. That’s where we come in: aligning organizations with proven, forward-thinking partners so security supports the business and evolves to keep up with attacks that become more sophisticated every day. This is another example of how we do Tech ONST-ly!

I’ve been noticing a clear shift in recent security and enterprise AI coverage. AI agents are no longer “tools.” They’re being treated as a new security perimeter. Once software can investigate, triage, and act, the hard problems stop being about intelligence and start being about identity, authorization, and auditability. At the same time, security teams are under pressure. Headcount isn’t growing, budgets are tight, and AI is being asked to take on more responsibility. But one rule keeps surfacing: automation only scales if accountability stays human. The real value of agentic AI will come from decisions that can be audited, especially in tightly controlled or disconnected environments. If you can’t explain why an agent acted, who authorized it, and where it ran - you’re not ready to deploy it. #AgenticAI #EnterpriseAI #AIGovernance #Cybersecurity #SecureAI