News and Analysis

As the old saying goes, sometimes “you have to fight fire with fire”. That’s certainly true in 2026, with cyber threats increasingly coming from ever more sophisticated use of AI tools. There’s a problem with that, though: most businesses are not yet fully set up to deal with AI-based attacks and don’t trust the tools to do the job. A 2026 Proofpoint AI and Human Risk Landscape Report revealed that half of organizations using AI-based security controls still experienced suspicious or confirmed AI-related incidents. There is a fundamental security gap in many organizations: AI adoption has outpaced the right security measures. In the report, nearly 9 in 10 (87%) organizations had moved AI assistants beyond the pilot stage, and 76% were actively piloting or rolling out autonomous agents. But that activity has outpaced security maturity: 63% had AI security controls in place, but 52% weren’t completely confident those controls would detect a compromised AI. AI tools are embedded directly into communications to increase productivity and speed, but in the rush to be efficient and compete, AI permissions and access to sensitive data are being left unchecked. These aren’t complex security engineering problems: they are organizational and process gaps that can be addressed without waiting for the tooling market to mature. Currently, only a few organizations have developed their incident response playbooks, logging coverage, or forensic tools needed to investigate a compromised AI agent. For many businesses lacking the in-house skills needed to take on these items, which can be significant, contracting a trusted managed service provider (MSP) can help bridge the gap. Many MSPs have been assisting businesses with AI adoption for some time now, and many have the deep understanding and technical skills needed to help businesses of all sizes see tangible benefits to AI while keeping critical data safe.The origins of AI attacks Most attacks start with unrestrained access and end with autonomous systems exposing sensitive data from these environments. When threat actors target agentic systems that lack proper controls, they don’t need to trick employees to access internal intelligence; they only need to manipulate the AI. Prompt injection attacks are a common way to do this. A bad actor might send a target user seemingly helpful AI instructions while posing as a trusted authority or co-worker. A well-intentioned employee may then ask an AI to answer what seems like a simple inquiry. Depending on the attacker’s instructions, the AI agent may instead be tricked into reading manipulated webpages (i.e., white text on a white background) to unwittingly extract internal data and send it to the attacker’s server.For prompt injection attacks, it’s important to limit AI agents’ access to only the tools and data they need to complete the designed task. This is a good solution to prevent AI from giving out more information than required. This can limit the scope of an external threat actor’s reach.That said, the employee’s role isn’t lost in all AI-based attacks. An ongoing cybersecurity skills gap severely impacts defenses, and threat actors know this. Over the last decade, multi-factor authentication (MFA) has been an important step toward stronger security authentication. But today, attackers can pair AI-generated phishing with ‘MFA bypass kits,’ such as open-source Evilginx (known as a penetration testing utility for these styles of attacks) and the W3LL panel (a private phishing kit) to deceive employees into handing over that ‘extra step’ of security. Tools like Evilginx and the W3LL phishing kit are used to create realistic sign-in pages that mimic those of Google, Microsoft, and others. Without proper security training, employees may unwittingly be signing into these while attackers capture their session tokens – even those with MFA. A good defense against MFA bypass kits is adopting phishing-resistant MFA technologies such as FIDO2 hardware keys, Windows Hello for Business, Certificate-based Authentication (CBA), and Passkeys. These methods are tied to legitimate sign-in pages and don’t work on fake pages. However, stopping the threat from ever materializing starts with having proper cybersecurity awareness training. Nearly half of all organizations lack this training or simply adopt a checkbox approach, which is why implementing these programs is an important step to closing the gap. These programs teach users to spot a myriad of cyber threats, including AI-based threats. Tools of this type are also a good example of using AI in defensive security, as some can leverage AI to customize the training an end user receives based on their performance in past training. This is another area where MSPs are highly qualified to assist. MSPs typically run training programs across a vast number of users and industry types. They understand what training works and what doesn’t, and can help position the best security awareness training for a given organization.How does AI enhance threat detection?To understand the power and importance of AI-powered cybersecurity, it helps to understand how it works. Once trained, a detection model becomes exceptionally good at spotting the characteristics of malicious activity. It can take into account thousands of different characteristics to spot anomalies, outliers, and similarities that humans are unable to correlate. It’s important to have a reliable cybersecurity service provider with REAL AI skills, because machine learning systems aren’t perfect and (while rare) can produce false positives. A strong partner can minimize these false positives while offering real-time threat detection and analysis, as well as faster, well-informed response times. A trusted MSP will have a wide range of capabilities and will have a deep understanding of the protection methods that work well within their target industries. By leveraging that deep knowledge from their partner MSPs, businesses will benefit from great protection, even with today’s AI-powered attacks.Preparedness in 2026 and beyondTo operate in this new era, businesses must treat every AI agent as a high-risk workload identity. In practice, this requires working with reputable MSPs to implement strict least-privilege access to avoid data leaks, constant monitoring to protect the integrity of the data, and comprehensive employee awareness training.