Your Network’s Front Door Is Wide Open to Cyber Threats
AI is turning trusted tools and AI systems into targets
For years, cyberattacks followed a familiar rhythm. Attackers sent a phishing email with an attachment, zip file, or malicious link. Then, they waited for an unsuspecting user to click it, allowing malware to slip in and give them access to the computer’s broader environment.
Key Takeaways
AI is making it easier for bad actors to use trusted tools, identities, workflows, and software to break into networks, stay hidden, and exploit systems.
As a result, cyberattacks today are faster, more convincing, and harder to detect.
Booz Allen’s DFIR team helps Fortune 50 companies and other organizations respond to active threats, strengthen zero trust defenses, secure AI systems, and pair AI-enabled tools with skilled analysts.
Today, that rhythm is changing. Attackers aren’t breaking in so much as “walking through the front door,” says Andrew Carr, managing director of global incident response at Booz Allen.
Equipped with powerful AI tools, attackers are executing sophisticated social engineering campaigns to steal login credentials. They’re rapidly sending fraudulent emails and convincingly impersonating employees. Once inside a network, they use the tools and systems organizations trust—internal email systems, remote IT tools, built-in software—to stay hidden.
“When an attacker is able to mimic employee behavior to sneak inside a network and blend in, it becomes much more difficult to identify the threat and stop it,” says Carr.
Carr is a leader within Booz Allen’s Digital Forensics and Incident Response (DFIR) team. This group of NSA-certified responders handles 1,000 cases every year for organizations ranging from Fortune 50 companies to small businesses. They bring together front-line incident response experience, threat intelligence, zero trust expertise, and AI-enabled tools to spot threats faster and neutralize them before the damage spreads. The team publishes a monthly report that tracks the number of cyberattacks by industry and type (e.g., email compromise, ransomware) as well as ransom demands of high-profile ransomware groups around the world.
From the front lines of cybersecurity, Carr and his team are seeing firsthand how AI accelerates the pace and complexity of cyberattacks. The technology has created a dangerous speed gap between how fast attackers are moving and how fast security teams can respond.
“It used to take a threat actor weeks to build up enough information to carry out a targeted phishing attack on an executive,” says Carr. “With AI, attacks move in minutes.”
Hiding in Plain Sight
For government agencies, healthcare systems, and other organizations, the growing speed of attacks and the shifts in how they’re unfolding are dangerous because the signs of a break-in are harder to spot. A bad actor may be able to disguise its attack to look like an employee logging in, a help desk request coming through, or a familiar software tool doing what it normally does.
Once inside, attackers can move quietly, study the environment, and look for weak spots. AI makes that easier, helping attackers gather information about an organization’s leaders, vendors, partnerships, and internal structure. It also helps write more believable messages, create fake invoices, and automate multiple steps that used to require additional time and more hackers.
“People are getting tricked every single day,” Carr says. “We’ve seen a single hacker execute a coordinated ransomware campaign that brought down an entire enterprise. No crew. No custom malware. Just one determined adversary exploiting an organization’s trusted tools.”
When AI Is the Vulnerability
At the same time, the AI systems that organizations are deploying in their own environments—chatbots, AI assistants, agentic tools, and models connected to sensitive data or business applications—are becoming targets themselves. An attacker may not need to break into a database directly if they can manipulate or misuse an AI system that already has access to it.
Carr describes one case involving an external-facing healthcare AI chatbot designed to help users understand medical information. Because the system had not been configured with the right permissions and controls, an attacker was able to expose a large amount of protected health information.
“Many organizations are really keen to get these AI tools in play really quickly, but they don’t necessarily know what they should be doing to secure them,” Carr says.
Even publicly accessible AI tools can introduce vulnerabilities. In their June incident response report, the DFIR team described an incident where unauthorized use of ChatGPT for work-related tasks heightened the risk of sensitive company data being exposed.
Inside the New Cyber Fight
Part of Booz Allen’s commercial cybersecurity group, the DFIR team applies decades of cyber expertise and deep knowledge of adversary behavior to contain threats, provide intelligence on bad actors, tune detection tools, govern remote access, and monitor the systems attackers increasingly try to abuse.
That includes helping organizations adopt zero trust principles. With zero trust, networks continuously verify users, devices, and applications and limit access to what is needed to reduce how far an attacker can move if one account or system is compromised. One example of Booz Allen’s zero trust work is Thunderdome, the nation’s largest zero trust program and one of the most advanced cybersecurity efforts in the federal government today.
As AI becomes part of the attack surface, the DFIR team is also helping organizations secure those systems before attackers exploit them. This entails testing AI for vulnerabilities using permission reviews, risk assessments, and red teaming—controlled tests where cyber experts act like adversaries to find weaknesses. Booz Allen’s agentic AI Vellox cyber product suite supports malware analysis, detection engineering, adversary emulation, continuous monitoring, and remediation—all informed by Booz Allen’s cyber tradecraft and models trained on real adversary behavior.
AI Finds Threats—People Stop Them
While AI can help defenders move faster, it does not replace experienced cyber teams. The same tools that flag unusual activity still need people who understand context: whether a log in is normal, whether a request makes sense, whether a system should be talking to another system, or whether an AI model has too much access to sensitive data.
That is where Booz Allen’s work is focused: fighting AI with AI by combining advanced tools with deep expertise to help organizations see what automated defenses can miss.
“You can put the best tool in place, but if nobody’s watching it, it’s not really going to do much for you,” Carr says. “As AI increases the speed of attacks, we are focused on making sure there are skilled analysts who deeply understand the context, the environment, and the threat.”
