AI Will Be Used Against You

1. Your Exposure, Their Ammunition

A threatening email quotes a password you really used. It claims the sender controls your webcam and demands payment. The password may be genuine. The claimed access may be fiction.

The UK's National Cyber Security Centre describes this intimidation technique in sextortion messages: an old password can come from an earlier breach and make the threat feel credible. Change the password anywhere it is still used and report the message, but do not treat the quoted password as proof that the sender currently controls your device. Read the NCSC guidance.

Same attack logic, different local bait. Choose a country:

AI will be used against you

Leaked information plus AI makes personalised attacks easier to prepare and sustain. AI can interpret messy records, connect details across sources, draft plausible messages, translate them and continue a conversation. AI reduces the effort needed to use exposed information. The requirements differ by attack: account takeover needs usable access, and payment fraud needs a payment path. Other harm can arise from disclosure alone, such as exposing private medical information or giving competitors confidential prices.

The message can be false while the details inside it are true. That is why exposure changes the attack.

2. What Attackers Can Obtain

People and organisations deliberately publish information: job roles, product announcements, social posts, investor communications and contact details. Attackers use that context. Cyber exposure begins with information disclosed without the subject's intention or control: breached records, malware logs, leaked documents, credentials, access offers and criminal discussions.

Voluntary and involuntary information combine. A public job title may identify who approves payments. A leaked mobile number gives the attacker a private route to that person. A stolen contract supplies the counterparty, amount and timing.

Five categories of exposed information

Three different observations

A seller claiming to hold company data establishes that the claim was made. A request to recruit an insider establishes hostile interest. An authorised comparison showing that a sample is genuine establishes disclosure of that sample. These findings can demand action, but they do not say the same thing.

3. Why Old Information Remains Useful

Information does not have one expiry date. Its value depends on how an attacker uses it.

Four clocks, not one

Sounds too academic? A credential discovered today may have been stolen in 2020. The finding is new to you. The theft is not necessarily new. The question is what the material can still do.

4. How Exposure Circulates

Exposure does not follow one mandatory pipeline. Material can move directly from a breach to extortion, from an infected device to account access, from a supplier to a private community, or from a paste site into a phishing campaign. A marketplace and AI may be involved. Neither is required.

Copies can move between sources and uses in either direction. Several routes can exist at the same time.

Finding the same listing in another channel proves circulation. It does not automatically prove a second incident or provide independent corroboration. Sector-wide reporting can tell an organisation what to prepare for; it cannot prove that the organisation itself is affected.

5. What AI Accelerates

AI changes the economics of using exposure. It can reduce the time and skill needed for several parts of an operation:

The asymmetry matters. An attacker can use machine time to prepare more variants. A defender still has to establish which observation is genuine, what it means and whether action is justified. AI reduces some attacker costs; it does not remove the bottlenecks of access, execution and getting a person or system to act.

6. When Exposure Becomes a Company Incident

An exposure finding may reveal information that has already been disclosed or a compromise that has already occurred. It may also identify conditions that could enable future harm. A serious finding can justify urgent investigation while its source, scope or current access remains uncertain. Employees, contractors and suppliers connect organisational systems to devices, identities and relationships outside the company's direct control.

This case connects the clocks. Credentials can be stolen years before they are used. It also shows why the company boundary includes contractors and other parties that use company identities. The exposure was not the whole incident. Continued credential validity and missing controls made later access possible.

7. The Uncomfortable Truth and the Response

You cannot reliably recall information that has already been copied and circulated.

You can change a password, revoke a session, remove an exposed service, interrupt a payment, warn an employee and correct a process. Those actions matter. They reduce particular risks and consequences. They do not reach every copy of the original password, contract, address, biometric, message or image.

Individuals can discover some exposure, change reused passwords, enable strong authentication, verify urgent requests through trusted channels, limit unnecessary public information and report fraud. They cannot see or control every breach, malware log, private community, supplier disclosure or copy made by somebody else. Organisations hold more of the relevant systems, evidence and authority, so they carry the larger monitoring and response responsibility.

From Signal to Action

Exposure intelligence must reach a person or process that can do something about it. Collected findings need context, a decision, an authorised response and evidence of what changed.

Specialist exposure-intelligence providers supply collected observations and analysis from breach material, malware logs and other relevant sources. Cyber Intelligence House (CIH), which I founded, is one example. An organisation may use an external provider, its own collection capability or a combination. The coverage and evidence behind each finding must remain explicit.

These are functions, not a compulsory shopping list. One provider or platform may perform several of them. A small organisation may use a security provider and a simple case record; a larger organisation may connect several specialist systems. Each handoff still needs a responsible owner.

AI may help interpret genuine findings or coordinate an approved response when it has the required data, tools and authority. Its model knowledge is not a substitute for current exposure collection, and a generated recommendation does not change an account, device or policy by itself.

Reduced risk is not erased exposure. Muting an alert does not remediate the issue underneath it.

8. Turn a Finding into a Decision

What was observed

What this establishes

• The old employee information still circulates, while the known password condition has been remediated.
• The marketplace post makes a current claim that Northstar contracts are for sale.
• The sample contains genuine confidential business material connected to Northstar and its supplier.

What remains unknown

• Whether the disclosure originated from Northstar, its supplier or another party.
• Whether the seller holds the full advertised collection.
• Whether the seller has current system access.
• Whether the old employee credential has any connection to the contract disclosure.

Credible consequences

A fraudster could impersonate the supplier and request that the scheduled deposit go to a different bank account, quoting the genuine purchase-order reference and payment date. Those details make the request fit an existing transaction. Completing the fraud would still require the changed payment instructions to pass the buyer's controls.

The confidential prices could also give another party a negotiating advantage without Northstar making a payment, clicking a link or responding to a message. The sample establishes disclosure of the information; the observations do not establish that payment fraud or commercial exploitation has occurred.

Proportionate response

• Preserve the post, sample, timestamps and collection context.
• Investigate the disclosure jointly with the supplier and determine where the contract was stored and accessed.
• Review relevant identity, document and transfer activity for evidence of unauthorised access.
• Brief the staff responsible for the supplier relationship and payments on the exposed order details. Require any changed payment instructions to be verified through established supplier contacts and the approved bank-detail change process.
• Retain the verified remediation status of the old password and keep its possible connection to the contract disclosure marked unknown.

Revoking access or correcting a process can reduce risk. It cannot remove copies of the contract already held elsewhere. The unresolved source, scope and access questions remain visible until evidence changes them.

Finding:
Person, organisation, supplier or system concerned:
Captured / circulated / discovered dates:
What is established:
Source claims:
Plausible attacker use:
Conditions required for that use:
Possible consequence:
Relevance / possible severity / confidence:
Unanswered questions:
Response owner:
Next action (proposed):
Action taken:
Result verified:
Remaining risk:
Evidence that would change the assessment:
What remains exposed after the response:

Sources and Related Guides

Primary sources used in this draft

• UK National Cyber Security Centre: Sextortion guidance - old passwords used in threatening messages.
• Hong Kong Government: Combating fraud involving deepfake - official account of the prerecorded HK$200 million conference fraud.
• FBI/IC3: Criminals use generative AI to facilitate financial fraud - text, image, audio and video fraud patterns.
• Mandiant: UNC5537 targets Snowflake customer instances - historical infostealer credentials, customer-account compromise, data sale and extortion.
• Heiding et al.: Evaluating automated spear phishing on human subjects - 101-participant click-through study.
• Microsoft: Protecting tokens in Microsoft Entra ID - replay, reauthentication, revocation and token-protection controls.

Further reading on response

• NIST: Incident response recommendations and considerations for cybersecurity risk management.
• Microsoft Sentinel: Automate responses with playbooks - an implementation example, not a product recommendation.

Related guides on this site

• AI Security Primer - identify what an AI system can know, reach and do, and decide which safeguards and evidence it needs.
• Building with AI Agents - turn an idea into independently checked work while the human controls purpose and consequential decisions.
• AI Bots as Managers - establish shared context, repeatable processes and working connections for bot-managed responsibilities under human authority.

Version history

This primer is versioned like software. Published versions receive a dated badge and frozen snapshot.

• 2026-09-05: current. Rebuilt around attacker utility, circulation, evidence interpretation, organisational response and verification. View snapshot →
• 2026-04-08: first edition. Introduced cyber exposure, attacker use of leaked information and the original exposure-monitoring model. View snapshot →

Dr Mikko S. Niemelä - 2026

Last updated: September 5, 2026