Duncan Greatwood, CEO, Xage Security
The democratization of AI makes for an exciting moment where rapid change feels possible, but it should also give security teams pause. By lowering barriers to entry and amplifying technological scale and speed, AI is not only empowering legitimate users but also enabling hackers, much like the rise of ready-baked “ransomware-as-a-service” tools that fueled an explosion of attacks organizations are still battling today.
That concern is no longer theoretical. A report from Anthropic details how AI systems are already being used to autonomously plan, conduct, and accelerate cyber-espionage campaigns, including reconnaissance, phishing optimization, and scalable covert operations. At the same time, OpenAI has warned that its next generation of models could pose a “high” cybersecurity risk, acknowledging that increasingly autonomous AI may significantly lower the barrier to sophisticated cyberattacks. Together, these developments confirm what many security leaders have long suspected: attackers are now using AI to exploit AI, widening security gaps that traditional defenses were never designed to handle. If history is any guide, the consequences of this new hacking accessibility will unfold for years to come.
Attackers are being emboldened by AI tools, which introduce a litany of new factors for security teams to consider. The breadth of their potential applications makes AI hacking tools both potent and dangerous, exposing previously unseen gaps in cyber defenses.
Tried-and-true Zero Trust security architectures have long been the gold standard for keeping organizations safe, and they are uniquely suited to this moment. AI-based threats can originate anywhere, and Zero Trust pillars such as identity verification, continuous re-authentication, and granular access controls will be vital to countering them.
Threat Actors Are Using AI to Hack AI
Rapid advancement in AI is arming threat actors with a robust suite of tools that enable intelligent attacks against organizations’ AI deployments. Whereas past tactics focused on breaching systems through technical vulnerabilities or stolen credentials, AI’s natural-language interfaces introduce a new attack surface that favors smooth-talking, relentlessly adaptive threat actors, whether human, AI-driven, or a combination of both.
Anthropic’s research underscores this shift, showing how AI systems can independently chain together actions by gathering intelligence, refining social-engineering tactics, and executing attacks at machine speed. OpenAI has echoed these concerns, noting that future models capable of operating autonomously for extended periods could dramatically expand attackers’ ability to probe systems, identify weaknesses, and sustain persistent campaigns. This marks a fundamental evolution in cyber risk: attackers no longer need deep technical expertise when AI can plan, iterate, and execute on their behalf.
With the growing power of large language models and “vibe coding,” technical skills are no longer a prerequisite for cybercrime. Some AI companies, including Anthropic and OpenAI, are investing in safeguards and governance, but the reality is that such protections are vulnerable to misuse and circumvention. Once bypassed, organizations are often left exposed without the visibility or controls needed to detect AI-driven abuse within their own environments.
According to IBM’s 2025 “Cost of a Data Breach Report,” 13% of organizations have reported an AI-driven breach, while another 8% cannot say for sure whether they’ve fallen victim to one. Perhaps most concerning, 97% of those breached reported having no AI access controls in place. This is not just irresponsible—it’s negligent—especially as CrowdStrike reports that 79% of detections in its “2025 Global Threat Report” were malware-free, signaling a shift toward hands-on, AI-assisted intrusion techniques.
Even job openings are now a point of vulnerability. North Korea has made coordinated efforts to secure American IT jobs using deepfakes and AI-driven impersonation to gain privileged access and steal sensitive data. Gartner expects this trend to accelerate, estimating that one quarter of job applicants will be fake by 2028. Even traditionally low-risk business processes like hiring must now be scrutinized in the AI era. Nothing is exempt.
Employees Are Using AI in Unintentionally Risky Ways
Despite pressure to integrate AI into daily workflows, employees remain uncertain about how to use it securely. A late-2024 CIO Dive report found that 56% of organizations lacked AI governance policies. Separate research from engineering hiring platform Howdy revealed that some employees feel pressured to use AI even when they’re uncomfortable, or simply pretend to use it to meet internal mandates.
All of this underscores the need for effective AI governance. That governance must be built on secure, user-friendly Zero Trust architectures that ensure only the right people and systems can access sensitive information, particularly as executives admit they may bypass internal protocols if AI makes tasks easier.
Zero Trust methods must account for human error and enforce safeguards by default, eliminating opportunities for misuse and protecting sensitive data across the organization.
Why Zero Trust Is the Answer
There’s a reason Zero Trust has remained the preeminent cybersecurity model even as threats evolve at breakneck speed. By assuming that threats are ever-present, organizations are less likely to overlook vulnerabilities and more likely to close gaps before they’re exploited.
It’s also cost-effective. A Gartner survey found that 78% of organizations that have adopted Zero Trust spend under 25% of their overall cybersecurity budgets on it, granting them leeway to add extra defenses on top of an already strong foundation.
Even the federal government has indicated that Zero Trust is the best way forward, with a Biden-era executive order leading to the creation of CISA’s Zero Trust Maturity Model and urging federal agencies to adopt Zero Trust principles.
AI-driven attacks exploit ambiguity and interpretation—areas where AI excels. Zero Trust restores clarity through strict authentication and authorization for every action, containing damage even when breaches occur. It provides a systematic framework capable of withstanding adaptive, AI-powered threats.
AI remains a massive opportunity despite the risks it introduces in the hands of attackers. Zero Trust principles offer a proven foundation for mitigating AI-on-AI hacking, allowing organizations to innovate confidently while protecting what matters most.
