As Anthropic’s Mythos mannequin makes clear, AI is a cybersecurity sport changer. When launched in preview, Mythos proved unexpectedly efficient to find and exploiting bugs in software program. Anthropic mentioned the preview launch discovered greater than 10,000 vital vulnerabilities throughout each OS and software it encountered. Along with exposing the bugs, Mythos was capable of execute an assault chain round each, together with the total assault lifecycle: reconnaissance, exploit and lateral motion.
Mythos is not alone. Different AIs have been proven to function autonomously to generate assaults. The Jadepuffer assault found in early July 2026, for instance, executed a whole ransomware and extortion operation pushed end-to-end by an LLM — no human required.
AI-enabled assaults
It is clear that AI poses an unprecedented risk to cybersecurity. Nevertheless it’s essential to notice that AI-enabled assaults aren’t a separate class of assault; they’re classical assaults at unprecedented velocity, scale and mutability. They embody all the usual kinds of cyberattacks, comparable to the next:
Like organic viruses, AI-enabled assaults mutate, and accomplish that at warp velocity. This makes response much more tough, since human defenders reply at human velocity. Because of this, AI-enabled assaults have change into a precedence for CISOs. A current survey by Darktrace discovered that 78% of CISOs now say AI-powered cyberattacks are having a major affect on their organizations.
These assaults aren’t hypothetical. Actual-world incidents comparable to the next happen repeatedly:
- AI-driven phishing. Microsoft risk groups flagged phishing assaults through which AI crafts phishing lures that extra simply evade detection by seeming extra human. These assaults embody practical personalization and imitation of human grammar and speech.
- Deepfake social media scams. A lot of social media is now AI-generated, together with propaganda meant to inflame passions, unfold lies or solicit cash.
- AI-enabled malicious insiders. North Korean operatives use AI to illegally land jobs with U.S. firms, placing enterprise knowledge safety at vital threat.
Behavioral biometrics
CISOs and their groups may stymie many AI-driven assaults if they might decide whether or not an attacker is an AI agent or a human — an enterprise model of the Turing Check. That is the main target of the burgeoning discipline of behavioral biometrics. Behavioral biometrics analyze person exercise to differentiate people from AI brokers. These applied sciences work passively within the background, constantly monitoring behavioral parameters through the course of a person session and studying as baseline norms evolve.
In contrast to elementary humanity checks comparable to CAPTCHA, which affirm human identification solely at the beginning of a session, behavioral biometrics can detect whether or not an entity that beforehand handed as human is failing to behave as anticipated, thus indicating a doable session hijack or a classy AI agent.
One other noteworthy element of behavioral biometric applied sciences is that they don’t seem to be static. Bodily biometrics comparable to fingerprints or retinal prints are distinctive to a particular human being and constant throughout the lifetime of that human being. Behavioral biometrics, in distinction, have a look at the patterns of interplay people have with know-how. These are discovered habits that may fluctuate over time and apply to greater than a single individual.
Conduct biometrics parameters
Most behavioral biometric techniques have a look at a number of parameters. These embody however should not restricted to:
- Typing patterns. Contains typing velocity, typing rhythm, keystroke strain, dwell time on every key and flight time between keystrokes. People are naturally erratic whereas AI brokers are unnaturally exact.
- Mouse motion patterns. Contains velocity, acceleration and cursor positioning. Once more, human customers naturally hesitate and proper mouse actions, whereas AI brokers carry out with machine precision.
- Touchscreen gestures. Present how customers swipe, scroll and faucet on cell units. These actions can exhibit distinctive interplay types particular to people. Equally, behavioral biometric techniques can monitor how customers maintain and orient their cell units whereas utilizing them. By capturing accelerometer and gyroscope knowledge, these techniques can distinguish between human and artificial machine actions.
- Navigation evaluation. Tracks how customers transfer by means of purposes and web sites and full numerous duties, together with web page go to length, activity sequencing and type completion.
- Gesture and motion evaluation. Tracks how customers alter posture and motion whereas working. This will embody hand positioning, head motion and eye monitoring. Gesture recognition analyzes intentional hand and finger actions customers make throughout video interactions.
These parameters could be mixed with a just about infinite array of different parameters to verify {that a} given person is, actually, human and that they’re the actual licensed human. Different parameters would possibly embody geographic location, IP deal with and machine kind. For example, an alert would set off if an worker in New York who has by no means logged in on a cell machine all of a sudden exhibits up on a telephone in Los Angeles.
How behavioral biometric techniques work
The fantastic thing about behavioral biometric techniques is that they work within the background, with out interfering with the person expertise. The techniques routinely seize machine data comparable to keystroke timing, mouse coordinates, touchscreen strain and machine place. A centralized system then analyzes the collected knowledge to construct complete behavioral profiles within the mixture and for particular person customers, identifies patterns, establishes baselines and calculates regular conduct ranges. Consistently evaluating person conduct in opposition to established profiles, the system then flags deviations from established patterns and triggers responses starting from contacting the person to blocking the session.
What’s subsequent for safety groups?
The early use of behavioral biometrics exhibits promising outcomes. Within the monetary sector, Mastercard’s 2025 cost fraud prevention analysis discovered that 42% of issuers saved greater than $5 million in fraud makes an attempt over two years utilizing behavioral biometrics.
In e-commerce and on-line retail, organizations already use behavioral biometrics to establish bot assaults, credential stuffing and account sharing, in addition to to stop stock hoarding, coupon abuse and cost fraud. Manufacturing firms and different organizations with high-value mental property use behavioral biometrics to guard amenities and detect and defend in opposition to AI-enabled insider threats.
What can CISOs do to implement behavioral biometrics to guard in opposition to AI-enabled assaults in their very own organizations? Listed below are 4 suggestions:
- Choose particular use circumstances with clear quantitative metrics and targets.
- Choose a set of parameters to implement initially, with others to return later.
- Assess correctly, searching for techniques that target your specific use case and combine together with your current and future techniques. You may reduce human effort by integrating into current dashboards and automatic techniques.
- Constantly monitor and tweak behavioral biometrics instruments to optimize outcomes, remembering that these aren’t set-it-and-forget-it applied sciences.
Johna Until Johnson is CEO and founding father of Nemertes Analysis, the place she units analysis course and works with strategic purchasers.
