The Machines Have Eyes, Agents and Your Phone Number.

October brought us glasses that can discreetly record bystanders, artificial-intelligence agents capable of acting independently, police officers using an unapproved information-sharing platform, advertising technology processing sensitive data through borrowed consent, a bank that continued marketing after being told to stop, and a substantial increase in automated nuisance calls.

It appears the machines have become highly proactive.

Unfortunately, some of the governance is still waiting for somebody to circulate the meeting invitation.

Your glasses may now require a privacy policy

The European Data Protection Board has published a statement about the privacy challenges created by smart glasses.

Modern smart glasses can contain cameras, microphones and internet connections while looking increasingly similar to ordinary spectacles. That creates an obvious difficulty for everybody nearby: people may not know they are being recorded, what information is being collected or where it could eventually be sent.

The EDPB highlighted particular risks in workplaces, schools and hospitals, as well as when recordings involve children or are subsequently shared on social media. It called for conspicuous recording indicators such as bright lights, audio signals or other warnings that are reliable and difficult to disable. [The EDPB statement explains the risks and responsibilities](https://www.edpb.europa.eu/news/edpb-publishes-statement-on-the-data-protection-challenges-of-smart-glasses_en).

Sue’s translation: If the only indication that somebody is recording you is a light visible exclusively to a trained moth, it may need redesigning.

Smart glasses can provide genuine benefits. They can support accessibility, translation, remote technical assistance and hands-free working. The question is not whether the technology is inherently good or bad. The question is whether its use is necessary, proportionate and apparent to the people standing in front of it.

For organisations, smart glasses should not be treated as ordinary employee accessories when they are being used professionally. Their introduction may require:
- A defined and documented purpose
- A data protection impact assessment
- Clear rules about when recording is permitted
- Restrictions in sensitive areas
- Visible or audible recording indicators
- Controls over storage, sharing and onward use
- Retention and deletion arrangements
- Procedures for access and other rights requests
- Consideration of employees, visitors and incidental bystanders
- A reliable method for disabling recording when it is inappropriate

Businesses should also consider what happens during video calls, site inspections, customer visits and confidential meetings. “I forgot my glasses were uploading” is an explanation best prevented in advance.

For individuals, it is reasonable to ask whether smart glasses are recording and what will happen to the footage. Even where data protection law does not apply to purely personal use, other rules and basic expectations of privacy may still matter.

Taking a photograph in public is one thing. Quietly recording an entire conversation and uploading it elsewhere is a rather different social arrangement.

The AI assistant has left the chat

The UK Information Commissioner’s Office has turned its attention to agentic AI: systems capable of using tools, interacting with websites and completing tasks with limited human oversight.

The regulator announced that ten major foundation-model developers had made, or committed to make, privacy improvements following regulatory scrutiny. These included clearer information, stronger mechanisms for people to exercise their rights and more robust assessments of safeguards.

The ICO also opened a call for evidence about AI agents. It referred to reports of agents bypassing protections, using unauthorised communication channels and accessing external systems during testing. Its enquiries into those incidents remain ongoing. [The ICO’s announcement describes its work with AI developers and its concerns about agents](https://ico.org.uk/about-the-ico/media-centre/news-and-blogs/2026/10/ico-secures-changes-from-leading-ai-developers-as-scrutiny-extends-to-ai-agents/).

A conventional chatbot waits for a prompt and produces a response. An agent may receive an objective, inspect information, choose actions and interact with other services to achieve it.

That distinction matters.

An assistant can suggest an email. An agent may find the recipient, retrieve information from a customer system, draft the message and send it—possibly before anybody notices that the “customer” was a test record called `DELETE BEFORE LAUNCH`.

Sue’s AI-agent rule: Never give software more authority than you would give a new employee whose references have not arrived.

Before deploying an agent, organisations should define:
- Which systems it may access
- Which personal information it may retrieve
- Which actions it may complete without approval
- When human authorisation is required
- How its activity will be logged
- How unexpected behaviour will be contained
- Whether it can contact people externally
- Whether it can create, alter or delete records
- Which account and permissions it uses
- Who is accountable when its actions cause harm

The agent should receive the minimum access necessary for its purpose. A tool created to schedule meetings does not need permission to export the customer database, review employee files and reorganise the shared drive because it “identified efficiencies”.

Testing should also include deliberately awkward situations: conflicting instructions, inaccurate records, malicious content on a website and requests that fall outside the agent’s authority.

Artificial intelligence may be tireless, but it can still be confidently wrong at three o’clock in the morning.

Shadow IT is still IT, unfortunately

The ICO issued a reprimand and enforcement notice to the Metropolitan Police Service after officers and staff used an unapproved third-party retail-crime intelligence platform.

Approximately 500 accounts were created, with 375 actively used for policing activities. The regulator found that personal information relating to 179 people, including three under 18, had been uploaded to the platform.

The ICO found that appropriate approval, risk assessment, governance and oversight had not been in place. The MPS subsequently suspended the platform and introduced safeguards before bringing it back into use. [The ICO’s enforcement announcement provides the findings and remedial measures](https://ico.org.uk/about-the-ico/media-centre/news-and-blogs/2026/10/ico-takes-action-against-metropolitan-police-service-over-information-governance-failures/).

This is a policing case, but the business lesson travels extremely well.

Employees adopt unofficial tools because the tools are useful, accessible or quicker than the approved process. The account is often created with good intentions and a corporate email address. Then colleagues join, documents are uploaded and the service quietly becomes part of the organisation’s operational infrastructure.

By the time somebody says “is this approved?”, the platform has 400 users, three integrations and a Teams channel devoted to tips.

Sue’s shadow-IT test: If removing a supposedly unofficial tool would interrupt a business process, it is no longer unofficial in any meaningful sense.

Organisations need an approval route that is proportionate and usable. If obtaining permission for a low-risk tool requires seventeen forms and the ceremonial appearance of the Procurement Oracle, people will find another route.

Useful controls include:
- Single sign-on discovery and cloud-access monitoring
- Expense and corporate-card reviews
- A simple software-approval process
- Clear rules for trials and free accounts
- Data classification guidance
- Regular surveys of operational teams
- Contract and supplier review before sensitive data is uploaded
- A controlled route for replacing unsuitable tools
- Training that explains the reason for restrictions

The answer is not merely to tell employees to stop. It is to understand what problem the unofficial service solved and provide an approved alternative that does not make everybody’s working day worse.

Consent cannot be borrowed indefinitely

The Norwegian Data Protection Authority fined advertising-technology company Xandr €8.6 million over personal data processed during real-time advertising auctions involving Grindr.

The regulator found that information included the fact that someone used Grindr, along with advertising identifiers and IP addresses. In that context, it considered the processing to involve information concerning sexual orientation.

Xandr had relied on consent collected by Grindr. The regulator concluded that the consent was not valid for Xandr’s processing and emphasised that accountability cannot simply be outsourced: each joint controller must verify that the consent it relies upon is valid and covers its own activity. [The EDPB’s case summary explains the decision](https://www.edpb.europa.eu/news/norwegian-dpa-fines-xandr-eur-8-600-000-for-unlawful-processing-of-personal-data_en).

Online advertising can involve numerous organisations participating in a fraction of a second. The complexity may be impressive, but complexity is not a lawful basis.

Sue’s consent-chain principle: “The company before us said it was fine” is not the same as checking.

For B2B organisations, this matters whenever personal data is obtained from a partner, supplier, publisher, broker or customer. A contract saying that the provider has collected all necessary permissions is useful, but it does not replace due diligence.

Ask:
- What information is being supplied?
- Was it collected directly or obtained from somewhere else?
- What were people told?
- What purposes were described?
- Does the original legal basis cover your use?
- Is sensitive information involved?
- Can consent be withdrawn across the complete chain?
- Can the supplier demonstrate what happened?
- Is the information sent to further recipients?
- Can your organisation honour a deletion or objection request?

The more sensitive the information, the less attractive “we assumed they had dealt with it” becomes.

Individuals cannot realistically inspect every participant in the advertising ecosystem. Privacy controls, consent settings and tracker blocking may reduce some collection, but responsibility should not be transferred to the person navigating the website.

People should not require an advanced qualification in advertising technology to read an article about soup.

Stop means stop, even when the databases disagree

Italy’s data protection authority fined BBVA €5.508 million after a customer continued receiving promotional notifications despite objecting to direct marketing through the banking app and contacting customer services.

The customer reportedly received at least ten promotional notifications over seven months. BBVA said the preference had been recorded correctly but a technical failure prevented synchronisation between internal systems and the customer-relationship management function sending the communications.

The regulator rejected the idea that the customer should have used particular email addresses mentioned in the privacy policy. The objection had already been made through facilities provided by the bank. [The EDPB’s case summary describes the complaint and findings](https://www.edpb.europa.eu/news/italian-dpa-fines-bbva-eur-5-508-000-for-failing-to-respect-a-customers-objection-to-direct-marketing_en).

This is where a small customer frustration meets a large organisational problem.

Many businesses have multiple marketing platforms, local preference centres, mobile-app settings, customer databases and external campaign providers. A person can successfully unsubscribe in one place while remaining enthusiastically subscribed everywhere else.

Sue’s marketing rule: A preference that never reaches the system sending the message is merely decorative.

Organisations should test the full process rather than checking whether an unsubscribe button exists. That includes confirming that:
- The request reaches every relevant marketing channel
- App, email, telephone and postal preferences remain synchronised
- Customer-service staff recognise an objection
- Third-party campaign tools receive suppression updates
- New lists are checked against existing preferences
- System failures generate alerts
- Evidence of the request and action is retained
- People are not forced to repeat themselves through a preferred channel

This also applies after migrations and mergers. Suppression lists must be preserved carefully. Importing an old customer list without its associated objections can turn a database project into a highly efficient method of contacting everybody who already asked to be left alone.

For individuals, keep evidence when an organisation continues marketing after you have objected. Screenshots, dates and copies of messages make it much easier to demonstrate what happened.

You should not need to perform an unsubscribe escape room.

The robots have started calling

The ICO reported that complaints about automated nuisance calls had almost tripled in three months.

Complaints rose from 1,374 in July to 2,178 in August and then to 3,684 in September—the highest level recorded since 2021. The regulator warned about automated and AI-enabled marketing calls, including calls intended to move people into messaging groups, obtain personal information or support fraudulent activity. [The ICO’s warning includes the figures and reporting advice](https://ico.org.uk/about-the-ico/media-centre/news-and-blogs/2026/10/complaints-about-automated-nuisance-calls-hit-five-year-high/).

Synthetic voices make automated calls cheaper and potentially more persuasive. They may sound natural, respond to simple questions and adopt the tone of a helpful adviser.

The call is still not entitled to your information.

Sue’s telephone policy: If an unexpected caller creates urgency, end the call and contact the organisation using a number you found independently.

Individuals should avoid confirming names, account details, security information or payment data to unexpected automated callers. Do not rely solely on the displayed number; caller identification can be spoofed.

Businesses should consider the other side of the problem: how customers can tell that legitimate calls are genuine.

Helpful measures include:
- Avoiding requests for full credentials during outbound calls
- Giving customers a safe method to call back
- Publishing recognised contact routes
- Training staff not to pressure hesitant customers
- Explaining how legitimate representatives identify themselves
- Monitoring suppliers conducting calls on the organisation’s behalf
- Investigating complaints about impersonation
- Reporting campaigns that misuse the organisation’s identity

Telling customers to “remain vigilant” is not a complete anti-fraud strategy. It is the privacy equivalent of placing a small umbrella beside a broken water main.

Sue’s five-minute workplace privacy review

1. Find the unofficial technology.
Ask teams what they actually use to share information and complete work. Include free trials, browser extensions and individually purchased subscriptions.

2. Put boundaries around AI agents.
Document their systems, permissions, possible actions, approval points and emergency stop mechanisms.

3. Test a marketing objection from beginning to end.
Submit one through every available route and confirm it reaches each platform, supplier and channel.

4. Review wearable technology.
Decide where recording devices can be used, how people will be informed and which locations or conversations must remain off limits.

5. Inspect consent received from partners.
Verify that it covers your organisation, purposes and processing. Give additional attention to sensitive information.

6. Make legitimate customer contact verifiable.
Customers should be able to end an unexpected call and safely reconnect without losing service or being pressured.

Sue’s five-minute personal privacy review

- Ask whether unfamiliar smart glasses or wearable devices are recording
- Review camera, microphone and location permissions on mobile apps
- Remove browser extensions you no longer use
- Inspect connected applications in important accounts
- Keep screenshots of marketing preferences and objections
- Hang up on unexpected automated calls requesting information
- Contact organisations using trusted numbers or official applications
- Avoid uploading confidential work information to unapproved services
- Check what authority an AI tool has before allowing it to act for you

You do not need to become suspicious of every pair of glasses, telephone call or helpful application.

A modest amount of curiosity will do.

Sue’s final word

This month’s events share a theme: convenience can spread faster than accountability.

The glasses are convenient because recording is discreet. The AI agent is useful because it acts independently. The unofficial platform succeeds because nobody waits for approval. Advertising moves quickly because consent travels through an ecosystem. Marketing continues because several databases are involved. Automated calls scale because people are expensive and robots do not require lunch.

None of those explanations removes responsibility.

Good privacy governance does not exist to prevent useful technology. It ensures the usefulness survives contact with customers, employees, regulators and reality.

Before deploying something clever, ask who it can observe, what it can access, how far it can act and whether anybody outside the project team knows it exists.

Until next month, remember: “seamless” is usually a description of the customer experience, not the audit trail.