Fortifying Against AI-Enhanced Social Engineering Attacks

Jennifer Post

|

November 18, 2026

Social engineering attacks through email phishing and credential compromise have long been a cybersecurity concern for organizations and threat actors are now using AI to make these attacks more effective and impactful. According to security awareness training firm HoxHunt’s recent Phishing Trends Report, 40% of all reported attacks across HoxHunt’s global threat detection network in January 2026 were AI-generated phishing attacks that bypassed email filters. Cyber risk and insurance firm Resilience’s 2026 Midyear Cyber Risk Report also found that attackers are using AI to sharpen social engineering attacks and make them more convincing. This is a key factor fueling a spike in cyber losses this year. Resilience found 85.3% of incurred cyber losses in 2026 traced back to “a person believing a voice, a message or a request that looked legitimate,” up from 17.7% two years ago. These trends have implications for how companies need to address social engineering risks moving forward, both in terms of insurance coverage and practical mitigation steps.

The Insurance Industry Response

The increase in AI-enhanced social engineering attacks is forcing insurers to reevaluate their approach to covering the cyberrisk. “Insurance applications inquire specifically about cyber hygiene more than ever,” said Kevin Szczepanski, partner at Barclay Damon and co-chair of the firm’s data security and technology team. “You should expect increasingly detailed questions about your organization’s physical, electronic and administrative safeguards.”

Frank Costa, principal and national growth leader at World Insurance Associates, noted that social engineering losses now drive cyber claims. In fact, Coalition’s 2026 Cyber Claims Report found that 71% of funds transfer fraud claims came directly from social engineering. Costa said the market is splitting, and carriers are responding to social engineering in different ways.

  • Coverage terms are changing. Starting January 1, 2026, several carriers began excluding AI-generated deepfake fraud from standard social engineering coverage on renewal, according to Costa. Others did the opposite and rewrote their social engineering coverage to cover AI impersonation, including voice cloning and video deepfakes. Two companies with similar policies can now have very different protection against the same attack depending on their carrier.
  • Sublimits are the actual exposure. While social engineering attacks are increasingly a primary cause of loss, that is not reflected in the structure of some policies. As a result, organizations may not actually have the coverage they think they are paying for. “Social engineering is usually covered by an endorsement, not the main policy limit, and the endorsement limit is typically $100,000 to $250,000,” Costa said. “That was adequate when the typical fraud was one fake invoice. It is not when an AI-cloned CFO voice can authorize a seven-figure wire. Many businesses learn only after a loss that their multimillion-dollar cyber program pays a small fraction of that for social engineering.”
  • Coverage now depends on verification. More policies require the insureds to have followed specific procedures to prevent social engineering losses, such as verifying payment requests by calling colleagues back on a known phone number. If an employee skips that step, the claim can be denied. These procedures need to be updated to reflect the current risk. “A verification clause written for 2019 email fraud can void coverage in 2026 when the ‘verification call’ itself may be a deepfake,” Costa said.
  • The crime policy gap remains. Standard crime and fidelity policies exclude losses where an employee “voluntarily” sends the money, even when they were deceived. “That exclusion is the main reason so much deepfake fraud goes uninsured,” Costa said. “Without a specific social engineering endorsement, a business may have no coverage.”
  • Definitions are being rewritten. It is no longer enough to have cyberrisk policies written for just email since attacker are also now using voice, video, text and real-time messaging.

In order to ensure that coverage meets expectation, Costa advised that any CFO or risk manager ask their broker the following questions at their next renewal: Do we have an explicit social engineering insuring agreement? Does it name deepfakes and AI impersonation? Is the sublimit at least as large as the biggest single wire we send? What verification steps must we follow to get paid?

Practical Steps to Mitigate Social Engineering Losses

In addition to insurance coverage, organizations should consider the following strategies to protect themselves against increasingly sophisticated and convincing social engineering attacks, and to improve the odds of qualifying for cyber coverage in the event they must file a claim:

1. Take a Different Approach to Training

The days of spotting phishing emails through bad grammar and awkward language are gone. Businesses should look beyond simply training employees to identify phishing emails. “Look at the nature of the request. Is someone asking you to change payment details, enter your credentials, act immediately or skip a normal approval step?” Szczepanski said. “If so, verify the request using a phone number or contact method you already trust—not one in the message itself.” In a reflection of how AI is changing the threat, he said it is now critical to apply the same standard to voicemail and video as threat actors can impersonate voices and faces too. He advised verifying sensitive requests through a separate, trusted channel and practicing incident-response plans through an annual tabletop exercise.

2. Do Not Let One Click Become a Breach

“We spend a lot of time telling employees not to click suspicious links, but the reality is that someone eventually will,” said Danny Jenkins, CEO of ThreatLocker. “Good security is not about expecting people to be perfect—it is about putting controls in place so one mistake does not give an attacker keys to the entire organization.”

A single click should not become a company-wide crisis. Organizations can prevent that by implementing strong layers of protection to contain the risk of any compromise. “Strong sign-in safeguards, limits on employee access, alerts for suspicious activity and reliable backups can help contain the damage,” he said. “When an account is compromised, the company must also be able to shut it down quickly.”

3. Adopt a Default-Deny Approach

According to ThreatLocker's Cybersecurity Awareness Month guide, organizations should only allow trusted applications and processes to run and nothing else. “If an application or process does not have a legitimate reason to run, why should it be allowed to run?” Jenkins said. “A default-deny approach changes the equation from trying to identify everything that is malicious to only allowing what the business has already decided it trusts.”

A default-deny approach helps organizations maintain greater control over their environment, reduce risk and prevent unauthorized software from executing.

4. Enforce Least Privilege Access

Least privilege ensures users, applications and systems have access only to the resources required to do their jobs. “Least privilege is one of the simplest ways to limit the damage from a compromised account,” Jenkins said. “If a user, application or service does not need access to a particular resource to do its job, that access should not be there in the first place.”

5. Secure Remote Access

It is crucial to treat remote access as an attack path that needs to be controlled. “Remote access is essential for modern businesses, but every remote connection is another potential path into the environment,” Jenkins said. “Companies should be asking who can connect, what they can access, from which devices and whether that access is still necessary.”

Remote access should be tightly controlled. ThreatLocker recommends implementing MFA, role-based access controls and strict access policies, as well as closing inbound ports to help prevent unauthorized access while maintaining secure connectivity for remote users.

6. Reduce the Attack Surface

“Security teams cannot protect what they do not know they have,” Jenkins said. “Unused applications, outdated software, unnecessary services and forgotten remote-access tools create opportunities for attackers. One of the most practical things a business can do is regularly remove what it no longer needs.”

Network segmentation is another way to reduce what an attacker can access. According to ThreatLocker, network segmentation creates boundaries between systems, applications and resources, which helps prevent attackers from moving laterally across the environment and gaining access to additional assets.

“Clients who plan for the worst often find themselves strongest in the face of an attack,” Szczepanski said. “Good cyber hygiene is not just about buying another product; it starts with knowing your risks and implementing safeguards to mitigate them.”

Jennifer Post is an editor at Risk Management.