AI Will Be Used Against You
Introduction
Scams are getting dramatically better. AI writes grammatically flawless phishing emails in any language, clones voices in seconds, and generates fake video calls convincing enough to fool trained professionals. Every one of these attacks can be customised to you personally using data that has already been leaked about you.
This is the formula: your leaked data + AI = personalised attacks at scale. What used to require a skilled human con artist now runs 24/7 at zero marginal cost per victim.
This primer explains what cyber exposure is, how AI weaponises it, and what that means for organisations and individuals. If you have read the AI Security Primer, you know how AI systems can be attacked. This primer is the other side: how AI is used to attack you.
Table of Contents
1. What is Cyber Exposure?
Your press team manages your voluntary exposure to the world: press releases, marketing, and investor communications. That is controlled, intentional, and strategic.
Cyber exposure is involuntary disclosure. It is information about your organisation that is out there, not published by you, and not controlled by you. It includes data breaches, dark web marketplace listings, credentials harvested by malware, and forum discussions where criminals trade access to your network.
Cyber exposure is the attacker's view of you: what they can find, buy, or leverage without ever touching your systems directly.
How it differs from other security work
Cyber exposure shows what attackers can already find, buy, or use about you from the outside. Vulnerability scanning tests your systems from outside. Penetration testing simulates attacks against your defences. Compliance auditing checks processes and policies. Cyber exposure answers a different question: what is already exposed, circulating, or being traded about you.
The persistence problem
Exposure accumulates over time. Research on hacktivist campaigns found that when hacktivists leak data about companies, the data remains available long after the campaigns have ended. Other criminals reuse it because it saves enormous preparation time. A list of employees, IP addresses, API locations, and internal documentation can take weeks to assemble from scratch. When it comes from an old leak, attacking becomes dramatically cheaper.
Exposure accumulates. Each new leak adds to the existing picture, and old breaches are recombined with new data.
2. Exposure is Fuel for Attacks
The pipeline is straightforward. Your data gets stolen in a breach or harvested by malware. It appears on dark web marketplaces, sometimes for sale and sometimes shared freely. AI processes it by personalising messages, cloning voices, and generating fake identities. Then the attack hits you or your organisation, tailored specifically to what attackers know about you.
Every link in this chain has gotten faster and cheaper. Breaches happen at scale. Dark web marketplaces are as efficient as any e-commerce platform. AI has removed the last bottleneck: the human effort required to turn stolen data into a convincing attack. Attackers spend cheap tokens to generate endless variants. Defenders still spend human time and repeated verification to sort real from fake.
The following sections show what this looks like in practice.
3. AI Voice Cloning & Deepfakes
What is it
AI can convincingly clone many voices from short audio samples. Combined with deepfake video, attackers impersonate executives, family members, or colleagues in real-time calls. In short interactions, cloned voices can be difficult for listeners to distinguish from the real speaker.
Why it matters
Voice cloning turns publicly available audio and video into an attack tool. A conference recording, a podcast appearance, or a social media video gives attackers enough material to clone your voice convincingly. The more public presence you have, the more material they have to work with.
Real-world case
Arup $25M Video Call Fraud, Hong Kong, 2024
A finance worker at engineering firm Arup was tricked into wiring $25 million. Attackers used deepfake AI to impersonate the CFO and other senior executives on a multi-person video call. Every face on the call was fake. Every voice was fake. The employee had no reason to doubt what he saw and heard.
The Arup case is significant because it was not a crude trick. It was a real-time, multi-person video call where every participant was synthetically generated. The employee followed normal procedures. He verified the request visually and verbally, with people he recognised, in a meeting format he had used before.
The traditional advice, "verify the caller's identity," assumes you can trust your own senses. That assumption is now broken.
4. AI Phishing at Scale
What is it
AI writes polished, contextually aware phishing emails in any language. It scrapes your LinkedIn, GitHub, and breach data to reference real projects, colleagues, and habits. The broken-English warning sign is gone. So is the generic "Dear Customer" opening.
Why it matters
AI phishing is where cyber exposure becomes directly weaponised. The attacker does not need to guess what will be convincing. Your leaked data tells them. Your employer, your role, your recent projects, and your colleagues' names are all available from breaches, LinkedIn scrapes, and infostealer logs. AI assembles the pieces into a message that looks like it belongs in your inbox.
Real-world case
The Contract Redline Phish
An attacker learns from leaked emails, documents, or LinkedIn that a target is working with a real customer, vendor, or law firm. The email arrives with a routine business pretext: updated redlines attached, revised MSA for signature, final NDA comments, or an urgent DocuSign link before close of business. The link leads to a fake Microsoft 365, SharePoint, or DocuSign page. The victim thinks they are completing normal contract work, enters credentials, and hands the attacker access to mail, documents, and internal threads.
Cyber exposure turns phishing into normal-looking work. Real counterparties, document names, deal timing, and job roles give the attacker the context to send a message that fits naturally into the recipient's day.
AI phishing works in every language with equal fluency. An attacker in one country can target victims in dozens of others simultaneously, with messages that read like they were written by a native speaker who knows the target personally.
5. AI Social Engineering
What is it
AI chatbots can maintain long-term conversations, building trust over weeks or months. Deepfake video lets scammers appear on video calls. Pig butchering scams, where criminals build romantic or investment relationships before stealing money, now run with AI operators handling hundreds of victims simultaneously.
Why it matters
AI enables patient, personalised manipulation at scale. Romance scams, investment fraud, and business compromise can now feel genuinely human. Every detail the attacker knows about the target, from breached data, social media, or infostealer logs, makes the manipulation more convincing and harder to detect.
Real-world case
AI Virtual Kidnapping Scams, USA, 2024-2025
A pattern documented by the FBI: criminals call a parent and play an AI-cloned voice of their child crying, screaming, and claiming an accident or abduction. The parent sends thousands of dollars to a courier before discovering their child is safe. The cloned voice is generated from social media videos. Multiple cases across the United States have followed this exact pattern, with losses ranging from $5,000 to over $15,000 per victim.
These virtual kidnapping cases show how cyber exposure feeds social engineering directly. The attackers need three things: a target's phone number, knowledge that they have a child, and audio samples of the child's voice. All three are available from social media and data breaches. All of it is involuntary disclosure. AI turns that exposure into a weapon.
Social engineering has always exploited trust and urgency. AI just removed the scalability constraint. One attacker can now run thousands of personalised cons simultaneously, each one sustained for as long as needed.
6. What Gets Leaked: The Five Exposure Categories
When people hear "leaked data," they think of passwords. But cyber exposure is much broader than that. It falls into five categories, and each one enables a different kind of attack.
Credential exposure is the most immediately dangerous because it provides direct access. A leaked password or session cookie lets an attacker walk in through the front door. Personal information exposure enables social engineering. Addresses, hobbies, family details, routines, and biometrics help attackers build believable bait, and some of that data cannot be changed. Financial data exposure enables direct fraud. Internal document exposure reveals your strategy, your vulnerabilities, and your decision-making processes. Infrastructure exposure gives attackers a map of your network.
These categories are not independent. A single infostealer infection can produce exposure across all five simultaneously: credentials, personal details, financial data, internal documents accessed through corporate email, and infrastructure details from saved VPN configurations.
Exposure accumulates. Old breaches get recombined with new data. It never expires.
7. The Underground
Most people have never seen the places where leaked data moves around: dark web forums, deep web services, paste sites, Telegram channels, ransomware leak sites, and infostealer marketplaces. This is where exposure is copied, traded, discussed, and weaponised. Understanding that ecosystem helps you understand why exposure monitoring matters.
Data breaches
Real-time tracking of new breaches, leaked databases, and credential dumps. Billions of records are exposed every year, and breach compilations aggregate historical data from hundreds of separate incidents into searchable collections.
Dark web marketplaces
Structured commerce exists across dark web and deep web services, with vendor ratings, customer reviews, and escrow services just like any e-commerce platform. Credentials, network access, personal data, and exploits are bought and sold there. Initial access brokers specialise in selling verified entry points to corporate networks.
Infostealer malware
Lumma, Stealc, Redline, and other infostealer families harvest passwords, cookies, and autofill data from infected devices at massive scale. They are one of the largest sources of fresh credential exposure today. A single infection can capture every password saved in a browser, including corporate logins on personal devices.
Threat actor communications
Forums, Telegram groups, paste sites, and ransomware leak sites are where attacks are planned, discussed, signalled, and boasted about. A company named in these channels is already attracting attacker attention. That usually means more attempts, more scrutiny, and more effort.
What exposure monitoring means operationally
Exposure monitoring systematically collects data from all four source types, normalises it into a consistent format, correlates findings across sources, and prioritises what needs immediate action. A credential in an infostealer log matters more when the same organisation also appears in a forum discussion, a paste-site leak, or an access-sale listing.
8. The Credential Cascade
This is how one infected personal device compromises your entire organisation. It is one of the most common paths from exposure to breach.
Employee's personal laptop gets infected
Downloads a cracked application or clicks a malicious ad. Infostealer malware activates silently, with no warnings and no symptoms.
Malware harvests everything
Browser passwords, cookies, and autofill data, including corporate logins saved in Chrome. The employee may have "remember password" enabled for the company VPN, email, and cloud platforms.
Credentials appear on the dark web
Within hours, the stolen data is packaged and sold on Russian-language marketplaces. Your corporate credentials are now available to anyone willing to pay a few dollars.
Attacker logs into your corporate systems
Session cookies bypass MFA entirely. The attacker does not need the password or the second factor. They are inside your VPN, email, or cloud platforms with a valid, authenticated session.
AI customises the next attack
Internal emails, org charts, and financial data fuel a targeted business email compromise or ransomware campaign. The attacker knows your approval workflows, your vendors, and your payment processes.
The critical detail in this chain: the infection happened on a personal device that your IT team does not control. The employee used the same browser for personal browsing and corporate work. Your perimeter was never breached. Your credentials simply walked out the door.
Session cookies are particularly dangerous because they bypass multi-factor authentication. MFA protects the login process, but a stolen session cookie represents an already-completed login. The attacker inherits a fully authenticated session without ever seeing a password prompt.
9. Your Exposure, Their Ammunition
Every piece of leaked data enables a tailored scam. The attack logic is identical everywhere. Only the local references change.
Same attack patterns, different local bait. Choose a country:
| You bank with OCBC Phone number + banking relationship from a leaked customer record |
→ | "OCBC Fraud Alert: confirm this S$1,280 transfer now or your account will be temporarily locked." |
| Your kid goes to Raffles Family info from leaked school records |
→ | "Urgent: your child was involved in an incident at school. Call this number immediately." |
| You flew SQ to Tokyo last week Travel data from a breached booking platform |
→ | "Singapore Airlines: your flight SQ638 is eligible for a $180 refund. Claim here." |
| You see Dr Tan at Mt Elizabeth Medical visit data from a healthcare breach |
→ | "Your outstanding balance of $420 at Mt Elizabeth is overdue. Pay now to avoid referral." |
| You earn $14k/month at DBS Salary + employer from a breached HR platform |
→ | "DBS HR: your March payslip requires verification due to a CPF adjustment. Log in here." |
Every one of these messages references something real about the target. The bank relationship. The specific school. The exact flight number. The doctor's name and hospital. The employer and a plausible payroll issue. Some of those details may be old. That does not make them useless. Old leaked data still works as validation material. It helps the attacker sound legitimate enough that the victim believes the message is real.
Before AI, crafting these messages for each individual victim was too labour-intensive to be profitable. An attacker might personalise a handful of high-value targets. Now AI processes breach data and generates thousands of customised messages per hour, each one referencing real details about the recipient's life.
These messages work because they reference real details that sound legitimate to the recipient. They contain information that used to be private and now is not.
10. The Uncomfortable Truth
As an individual, there is very little you can do.
Your data has already been leaked. Here is what you specifically cannot do about it:
- You cannot un-leak data that is already circulating in breach compilations and infostealer logs
- You cannot monitor the underground because dark web forums, Telegram channels, and infostealer marketplaces require specialised access, language skills, and operational security
- You cannot stop criminals recombining old leaks with new data to build a more complete profile of you
- You cannot stop AI from weaponising what is already exposed about you, including your voice, your habits, your relationships, and your credentials
- You cannot solve cyber exposure as a private individual because the problem is organisational in scale and requires organisational tools to address
Cyber exposure is an organisational problem that requires organisational monitoring and response.
What companies can do
- Buy continuous dark web monitoring
- Run exposure assessments
- Deploy credential monitoring at scale
- Train employees on AI-enhanced threats
- Implement session token rotation
- Enforce device management policies
What individuals cannot do
- Un-leak data that is already out there
- Monitor dark web forums themselves
- Stop AI from using their exposed info
- Prevent scams that use real personal data
- Detect which of their data has been leaked
- Control what their employer's systems expose
This is an organisational responsibility, not a personal one. The tools, scale, and access required to monitor exposure across dark web forums, infostealer marketplaces, breach databases, and threat actor channels are not available to individuals. Companies that employ people have a duty to monitor what is exposed about their organisation and their employees, and to act on what they find.
What acting on it looks like
- Credential exposure: Force password resets, enable MFA, review access logs, rotate API keys, and check whether unauthorised access already happened
- Session cookie exposure: Invalidate active sessions, force re-authentication, review access logs, and check which devices and locations used the stolen sessions
- Personal data exposure: Alert affected employees, heighten monitoring for social engineering attempts, brief the security team, and prepare for follow-on scams that use old data as validation material
- Internal document exposure: Determine what decisions, projects, or counterparties the material reveals, assess whether legal, commercial, or strategic harm is likely, and monitor for phishing or fraud built from that context
- Infrastructure exposure: Patch or decommission exposed services, rotate affected credentials, update firewall rules, and investigate how the exposed information became public
- Company data for sale: Verify whether the seller has real material, identify which system or breach the data came from, investigate the original compromise path, and prepare incident response, legal, and communications coordination
- Access for sale or active targeting: Treat marketplace sales, hacker-for-hire requests, forum targeting, and paste-site appearances as operational signals. Elevate monitoring, brief the SOC, protect the named assets, and consider incident response immediately
Do not just patch the leak. Investigate how it happened. A leaked credential means asking where it was stored, how it was captured, and what else may have been taken at the same time.
11. What Exposure Monitoring Looks Like
An exposure assessment is a continuous process of watching what is said about you, sold about you, and leaked about you in places you would never normally look.
In practice, exposure monitoring involves:
- Defining your keywords and identifiers: domains, IP ranges, brand names, key personnel, and financial identifiers. These are the search terms used to find your data across underground sources.
- Continuous collection: automated and manual scanning of dark web forums, infostealer marketplaces, breach databases, paste sites, Telegram channels, and ransomware leak sites.
- Interpretation and validation: leaks and breach fragments rarely arrive with full context or instructions. A credential listing may be old, fresh, standalone, or part of a larger compromise. The task is to identify what the artifact is, where it came from, how old it is, and what it means operationally.
- Correlation and prioritisation: connecting findings across sources to build an accurate picture. A credential in an infostealer log becomes far more important when the same organisation appears on an access broker listing, a paste site, or a threat actor discussion. Multiple incomplete signals often need to be combined before the real incident becomes clear.
- Reporting and action: translating technical findings into decisions: which credentials to rotate, which employees to alert, which systems to investigate, and what to report to leadership.
In practice, this process needs an owner, a cadence, and a clear path to action. Exposure monitoring is typically owned by the threat intelligence or security operations team, with findings reviewed daily for critical items such as active credential sales and ransomware leak site mentions, and weekly for the broader exposure picture. Critical findings go directly to the SOC or incident response team. Strategic trends, such as changes in exposure over time and new threat actor interest in your sector, go to the CISO and, where appropriate, to the board.
12. Further Reading
Related primers on this site
- AI Security Primer: How AI systems themselves can be attacked. Covers prompt injection, jailbreaking, model theft, data poisoning, and defences. If attackers are using AI against you, you should also understand how to secure AI systems you operate.
- Build with Agents: A practical guide to building with AI agents. If your organisation is deploying agentic AI, understand the security implications of autonomous systems that take actions on your behalf.
Frameworks and standards
- MITRE ATT&CK: Adversarial tactics and techniques knowledge base
- NIST Cybersecurity Framework: Risk management and response guidance
- EU NIS2 Directive: Network and information security requirements
- GDPR: Data protection and breach notification requirements
- SEC Cybersecurity Disclosure Rules: Material incident reporting requirements
- DORA: Digital operational resilience for financial entities
Last updated: April 8, 2026