AI Security | Breaking Cybersecurity News | The Hacker News

Many AI systems depend on input signals that teams cannot fully inspect or explain. These opaque sources reduce visibility into the data paths that influence model behavior. Engineers lose provenance records, limiting the diagnosis of abnormal outputs. This complicates the work of security teams that need clear records of what influenced a model at any point in time. Verifiable search data offers a stable alternative. It gives teams an input they can examine, store, and reproduce in controlled conditions. Engineers can compare model behavior against information that was publicly accessible at the time a result was produced, rather than depend on hidden internal signals. This article outlines why verifiable search data gives AI and security teams the clarity required to maintain operational control. Why Traceability Matters in AI Systems Traceability lets teams follow an input from its origin through each processing step. When every stage can be inspected, engineers can review...

Artificial intelligence is everywhere now. From automated code completion to autonomous infrastructure management, AI tools and AI agents help DevOps speed up deployment cycles and change how development teams operate in general. At the same time, this rapid adoption of AI has created a reality that is hard for security teams to ignore: as with the growth of AI capability within the software development life cycle, the attack surface also grows. In 2025, there were 68 AI-related incidents recorded across major DevOps platforms according to the 2026 DevOps Threats Unwrapped Report . In the first half of 2026, the number of AI-related incidents visibly grew — research from GitProtect Lab tracked 84 AI-related incidents in six months alone. Thus, comparing the first half of 2026 to the same period in 2025 shows that AI-related incidents in development environments have nearly tripled. What do DevOps and DevSecOps say about AI incidents in general? According to GitProtect Lab 's surve...

A security leader at a global finance company told us recently that his team discovered three times more AI tools running in their environment than IT had approved. Nobody had smuggled them in. Employees had simply pointed agents at their work, and the agents brought their own tools with them. That conversation is not unusual. It is the conversation. Over the past year, in customer discussions across finance, healthcare, manufacturing, and government, the same four struggles come up so consistently that we have started treating them as the shape of the problem itself. Every company effectively hired a second workforce this year, human workers and agentic workers side by side, and the agentic workers never went through onboarding. No handbook, no scoped credentials, no acceptable-use policy they can actually read. Here is what teams are struggling with, what our research says about why, and what closing each gap actually requires.

We had an enterprise customer tell us their entire AI footprint was Copilot. That was the whole answer. One tool, one line item, done. We ran the first scan. Copilot wasn't even close to number one. Claude was. OpenAI came in second. Copilot was third. Nobody on the security team knew because Claude doesn't show up the way a SaaS app used to show up. There's no single login screen, no single admin console, no one place to look. That's the part most security teams miss. Claude isn't one surface. It's six. The six places Claude actually runs Claude Enterprise and Connected Apps. This is the surface everyone pictures: employees typing into Claude, OAuth'd into Google Drive, GitHub, Slack, and Jira, asking Claude to act on what's inside. The audit log shows that a connection happened. It does not show what got pulled into the prompt or what came back out. A finance analyst can drop a quarter of board materials into a conversation in ten second...

Security is fundamentally about knowing what a system is supposed to do, then catching it when it does something else. For software with deterministic execution paths, that is a tractable problem. For AI agents, it is not. An agent does not follow a fixed code path. It reasons toward a goal, selects tools based on that reasoning, and adapts its next action based on what those tools return. The same input can produce a different sequence of actions depending on context, session history, and what an external tool happened to return. Behavior is the attack surface, and behavior changes. Traditional proxies and AI firewalls were built to inspect content: what a user sent, what a model returned. Intent security asks a different question: is this agent doing what it was built to do, in this context, for this user, right now? Answering that requires building a behavioral baseline for every agent and measuring deviation from it continuously. That is the problem Lasso was built to solve....

When I work on an incident, the first question I ask is almost never "what malware ran." It's "whose credentials did it use, and what was that account allowed to touch." Nine times out of ten, the interesting part of the story isn't the exploit. It's the access. The exploit gets you in the door. The privilege is what lets you walk through the building. For thirty years, that question had a human-shaped answer. A person clicked something, a person got phished, a person reused a password, an admin left a service account sitting on a domain controller with a password from 2014. The identity at the center of the incident belonged to somebody with a badge and a manager. That assumption is quietly dying, and most enterprises have not adjusted their controls to match. The actor on your network is increasingly not a person at all. It is a workload, a script, a bot, an API (application programming interface) client, and now an AI agent that can reason, plan...

Attackers prefer the path of least resistance. Why break in when you can log in? That is what makes working credentials so valuable. A leaked password, token, or API key does more than reveal sensitive data; it offers a way in. No vulnerability or privilege escalation chain necessary. If the credential is valid, the attacker can simply use it, and the session will look like ordinary activity. This shifts the questions from "where can attackers break in?" to "where do usable keys tend to accumulate? Increasingly, the answer is the developer laptop: one of the most instrumented machines most companies own, and one of the easiest places for credentials to go unnoticed. The endpoint is watched for malware, behavior, posture, and configuration. But a valid plaintext credential is something else: a door an attacker may not need to force. A fully patched fleet can still have usable keys scattered across its devices, enough to turn one foothold into something much larger....

The most important cybersecurity impact of artificial intelligence is not that attackers can write better phishing emails or automate parts of their workflow. It is that AI is changing the speed, scale, and decision-making dynamics of cyberattacks.  That creates a problem many organizations have not yet fully confronted: most cyber governance and incident response models were designed for human-speed attacks.  For years, security teams operated under a familiar sequence. Detect suspicious activity. Investigate. Validate the threat. Escalate to leadership. Decide on containment. Communicate with stakeholders. That model still has value, but it assumes defenders have enough time to build confidence before taking material action.  AI-enabled attacks challenge that assumption.  Adversaries can now use AI to accelerate reconnaissance, generate highly personalized social engineering, modify malware, test payloads, summarize stolen data, identify vulnerabilities, a...

Modern attacks are not primarily defeating infrastructure. They are inheriting trust. Identity Did Not Become Important. It Became Infrastructure. Security teams still talk about identity as though it is one security discipline among many, sitting beside endpoint protection, cloud security, network defense, and vulnerability management. That framing no longer reflects how modern enterprises actually operate. Modern business environments run on identity, delegated trust, cloud roles, automation pipelines, APIs, machine permissions, and continuously exchanged credentials. Users authenticate into SaaS platforms that the organization does not own. Workloads assume permissions that nobody provisions manually. Services trust other services built across years of acquisitions, migrations, technical debt, and operational compromise. The enterprise is no longer running on infrastructure alone. It is running on identity. Attackers recognized this shift before many defenders did. That i...

Launching a ransomware attack used to take real effort. Now, thanks to AI, almost anyone can launch a sophisticated attack, which changes the game for everyone responsible for protecting businesses. Reconnaissance that once took hours now takes minutes. Phishing emails that used to require careful crafting can now be generated at scale and sent to hundreds of targets simultaneously. IBM's 2025 Cost of a Data Breach Report found that AI reduced the time required to create phishing emails from 16 hours to just 5 minutes. For MSPs managing dozens or hundreds of clients, and for internal IT teams holding the line across an entire organization, understanding how AI is changing ransomware is key to staying ahead of the threat and minimizing disruption when attacks occur. The attack that starts in the inbox Before attackers can encrypt files or demand a ransom, they first need a way into the organization. One of the easiest ways to get that access is by tricking someone into cli...

The conversation around Anthropic's Claude Mythos Preview has understandably centered on zero-days. If AI systems can identify and exploit vulnerabilities across every operating system and browser at scale, defenders have to assume that exploit timelines will keep compressing. But for CISOs, the harder question is how long exposed access credentials remain valid after defenders discover the exposure. Credentials determine how far an attacker can move, how long they can persist, and how difficult containment becomes. A vulnerability just gets them in the door. That gap between time-to-exploit and time-to-revoke is where many organizations are most exposed. GitGuardian's State of Secret Sprawl report shows 64% of valid secrets detected in 2022 were still active and exploitable four years later in an environment where exploitation now collapses to hours. Vulnerabilities get attackers in the door, but credentials decide how far they go. The Mythos-ready briefing , developed b...

The event that didn't exist At 2:14 p.m. on a Tuesday, an employee clicks a link. If you reconstruct the moment from your security stack, nothing happened. A browser process opened an HTTPS connection. The certificate was valid. The destination wasn't flagged. Traffic volume was unremarkable. No detections fired. Inside the browser session, a different story was unfolding. The page that loaded looked like a routine CAPTCHA with "verify you're human" framing, a prompt to complete a quick check to continue. The instructions told the user to press Windows+R, paste what had already been copied to their clipboard, and hit Enter. In the middle of a busy work day, they did. What they pasted was a shell script. It executed in the user's own context, with the user's own permissions, as a deliberate action the user performed with their own hands. Nothing about the browser session looked unusual. The page rendered normal web content. The clipboard write happene...

The transition to container-first infrastructure is complete, with microservices now powering production-critical workloads and driving digital innovation for most enterprises. While 100% of DevSecOps leaders view containerization as critical to their production strategy, this shift has been accompanied by a crisis in security frameworks. According to the ActiveState 2026 State of Vulnerability Management & Remediation Report , respondents' organizations faced a staggering 82% container breach rate over the past year. Many companies have tried to mitigate risk by "shifting left", empowering developers to build security into their code from the start while still leveraging containers and open-source software from public registries. But in 2026, the reality of shifting left has often meant shifting a mountain of undifferentiated remediation work (i.e., fixing someone else's code) onto already overextended engineering teams. How should Security Leaders think about container se...

In November 2025, Anthropic revealed a cyber espionage campaign dubbed GTG-1002, the first documented case of an AI agent orchestrating real-world intrusions with minimal human input. A Chinese state-sponsored group manipulated Anthropic's Claude Code assistant into executing about 80% of a multi-target hacking campaign autonomously. Instead of merely advising cybercriminals, the AI took control of key steps: reconnaissance, vulnerability discovery, exploitation, credential theft, and data exfiltration across dozens of organizations. The result was an operation running at machine tempo. Claude performed tasks in a fraction of the time a human team would need, even identifying sensitive databases and writing exploits in seconds. Figure 1: The distinct phases of the Claude cyberattack At the peak of the attack, the AI made thousands of requests (often several per second), an onslaught of activity impossible for humans to match. This speed and scale of automation is a game changer: a...

Employees are increasingly using personal AI tools, AI-powered extensions, and emerging agentic browsers to accelerate their work. But unlike sanctioned AI platforms, these tools operate inside the browser runtime, where neither CASBs, SWGs, EDRs, nor DLP solutions have visibility. This has quietly turned the browser into an unmanaged AI execution environment, giving way to a new threat known as shadow AI. Shadow AI isn't just the latest buzzword; it's a serious risk that leaves organizations vulnerable to data loss, cyberattacks, compliance violations, and more.  What is Shadow AI? Shadow AI refers to GenAI-powered tools, browser extensions, and browsers that workers use on their own, without any company vetting or guidance. Different from shadow IT, where unsanctioned apps or devices slip through the cracks, shadow AI lives directly in the browser.  For example, employees might use their personal Claude accounts to work with sensitive company data or work on important pr...

For most of its history, the Security Operations Center (SOC) has been a privilege of the few. Building one meant millions in technology spend and round-the-clock analyst coverage. Unsurprisingly, for years, SOCs were a privilege of the few -  large enterprises and organizations with high-risk profiles, where budgets and scale justified the investment. Everyone else was left with partial coverage or had to outsource. That reality is changing. AI has flipped the SOC equation. What was once out of reach for all but the largest enterprises is now accessible and affordable for nearly every company that needs one. The risk every company faces By now, almost any 9-year-old knows that cyberattacks threaten every company . It's no longer just banks and financial giants in the crosshairs. Over the past decade, cyberattacks have expanded into every sector, from e-commerce sites to research institutes to local hospitals. Recent data from the 'VikingCloud 2025 SMB Threat Landscape' repo...

At some point in the last decade, SIEMs turned into that one friend who always promises to help you move, then shows up late, eats all your pizza, and still expects gas money. They were supposed to deliver centralized visibility and faster investigations. Instead, most SOC teams ended up with endless alerts, eye-watering bills, and dashboards that look impressive on the big screen but don't actually stop attackers. So, how did we end up here? A short history: when SIEMs were actually useful Back when firewalls were still exciting, SIEMs solved a real problem: logs scattered everywhere, auditors breathing down your neck, and no way to answer "who logged into what, when?" Then came the "next-gen" era. Vendors promised smarter detection, correlations across your stack, and even a pinch of threat intel. The promise was fewer false positives and a faster response. But instead of taming noise, NG SIEMs just amplified it. It was like turning up the volume on a broken radio and calling ...