Frequently asked questions.

Blanket bans usually lead to shadow AI: employees then use private accounts and nobody is in control. Approved tools, clear rules and training work better.

That depends on the provider, the plan and your settings. Many free services analyse input, business plans usually do not. Check the contract and account settings.

In Germany, copyright protects personal intellectual creations by humans. Purely AI-generated content is therefore usually not protected. Results may also infringe the rights of others.

In certain cases, yes, for example deepfakes and AI texts on matters of public interest (Art. 50 AI Act). Details are on the AI regulation page.

As a rule, whoever uses the system is responsible. In addition, new product liability rules covering software and AI apply in the EU from the end of 2026. Details under regulation.

Not reliably as things stand. So the rule is: limit the risk instead of hoping. Minimal permissions, separating instructions from data, confirmation for risky actions and never executing output unchecked.

They prevent data from going to a provider and give you more control. In return, operation, updates and the supply chain are your responsibility. Security comes from operation, not from location alone.

Sometimes, but only reliably with procedures: call back, code word, cross-check via a second channel. Never rely on what you see or hear alone.

It builds on IT security (identities, network, data, monitoring) and adds AI-specific risks. Both belong together: IT security, Monitoring, NIS2.

The Model Context Protocol is an open standard that lets AI applications connect tools and data sources. In practice, every connected server is a decision of trust and should be checked like any other software.

Technically yes. Legally, the agent's actions are usually attributed to whoever uses it. Such actions need approvals and clear limits. Have them checked legally.

Yes, like other AI systems, according to their purpose: depending on the application, transparency, high-risk or general obligations apply. More under AI regulation.

With a low-risk, reversible task at level 1 or 2, separate accounts with minimal permissions and logging. Then expand step by step.

At work, yes, as a so-called deployer, especially through the AI literacy obligation, transparency rules and data protection. Purely private, non-professional use is exempt from most obligations.

Authorities can order measures and impose fines, in Germany mainly the Federal Network Agency. The amount depends on the type of violation and can reach millions.

Deepfakes, meaning realistic-looking, artificially generated or manipulated content, must be disclosed. Providers of generators must mark content in a machine-readable way. Exceptions apply in other cases, for example for obviously artistic works.

As of 5 November 2026. Laws change, especially in the USA and with EU standards. Have important decisions checked legally.

A fixed role is not mandatory everywhere, but it makes sense: someone should be responsible for the register, policy and training. This can also be an existing role in data protection or IT.

A ban alone is rarely enough. Offer an approved alternative and explain the rules. Otherwise employees find workarounds.

With a few metrics: share of approved tools, trained staff, reported incidents, quality of results and time saved.

AI systems are part of the IT landscape. Risk management, supply chain, incident handling and training under NIS2 apply to them too.

Yes, with care: do not enter sensitive data, check the settings and verify the results. There is no absolute privacy with online services.

Secure evidence (screenshots, messages), report the incident to the platform and the police, inform your bank and change passwords. For images used against your will, also contact counselling services.

Only with the consent of the people shown. Faces are personal and sometimes biometric data, and image rights apply here too.

That depends on the context: at school, university and work there are often labelling rules. If in doubt, disclose it.

EDR monitors individual endpoints, XDR also covers email, network and cloud. MDR is the service behind it: a team analyses the alerts and helps with the response.

Usually not its own. If nobody can watch around the clock, an MDR or SOC service is an option. The basics come first: MFA, updates, backups and an emergency plan.

That depends on your industry and company size. Have it checked specifically, for example with an assessment or through your trade association. The measures it contains are good practice anyway.

With these five: password manager, MFA, automatic updates, tested backup, a reporting channel for phishing. The basic check helps with the order.

Indirectly, yes. Affected companies must take the security of their supply chain into account and pass requirements on in contracts. Smaller suppliers therefore also receive questionnaires and requests for evidence.

It covers many technical and organisational requirements, but does not replace registration, reporting obligations and management responsibility.

Operators of critical facilities are also subject to stricter rules. Since 17 March 2026 the KRITIS Umbrella Act applies to physical protection, more under Protected infrastructure.

Yes. The EU Cyber Resilience Act brings obligations for manufacturers of connected products, including reporting obligations. Which rules apply to you depends on your business model.

Yes. It does not have to be thick. What matters are roles, contacts, first steps and priorities. It must also be available when systems fail, for example printed.

Authorities generally advise against it, because there is no guarantee of recovery and it finances the attackers. The decision belongs in the crisis team, with legal advice and involving law enforcement.

It can cushion costs and give access to experts. However, insurers set requirements for security measures and reporting channels that must be met in advance.

RTO is the maximum time a system may be down. RPO describes how much data loss, measured in time, is still acceptable. Both determine backup and recovery concepts.

It detects known malware, but not attacks that behave like normal use. EDR and central analysis close this gap.

There is no blanket number. Purpose, risk, requirements and data protection are decisive. Set the duration deliberately and document it.

A SIEM collects and correlates logs from any source. XDR brings closely integrated detection and response from endpoints, email, network and cloud. The two are often combined.

Technically yes, even with open tools. The effort lies in operation: maintaining rules, assessing alerts and being reachable around the clock.

For critical applications and data centres, where lateral movement in the network is particularly dangerous. For smaller environments, clean zone separation is often enough.

A classic VPN gives access to the network. Zero Trust access also checks each application and device. The two can be combined.

Good with WPA3 and sign-in by certificate or user account (802.1X). Shared passwords for everyone are the weaker solution.

The BSI IT-Grundschutz or ISO 27001. Both provide a framework that can be tailored to your own protection needs.

Law and regulation define this by sector and thresholds. Operators are registered. Have your classification checked by experts, ideally with your trade association or specialist advice.

Not to the depth of a data centre. But server room, access, fire protection and backup power make sense for small businesses too and can be implemented with manageable effort.

It means current, proven procedures and standards available on the market. The level keeps evolving, so regular review is part of it.

An access event can display or store a camera image. This helps with alarms and investigations, but needs a data protection and permissions concept.

Local control is more independent of the manufacturer and protects data better. Cloud services are convenient but need a well-secured account. The two can be mixed.

Not necessarily. A guest network on the router is a good start. A VLAN gives more control, for example so that devices may only talk to certain destinations.

The risk lies less in the price than in missing updates and default logins. Look for update commitments, encrypted connections and the ability to change passwords.

Matter uses encrypted, local communication and works across manufacturers. But it does not replace updates or a separate network.

On your own property often yes, if purpose and scope are appropriate and public space and neighbours are not captured. A documented case-by-case assessment is decisive.

72 hours is the guideline. Longer is possible if there is a clear reason, for example weekends or rarely used rooms.

Dummies can also create surveillance pressure and are sometimes treated similarly by authorities and courts. Better to clarify beforehand.

Probably not. Audio recording is legally particularly critical and usually not permitted. If in doubt, turn it off.

The credential is blocked in the system without replacing a lock. With offline systems the block may take effect with a delay, so a clear process is important.

Offline is affordable and quick to retrofit. Online offers real time and central control. Wireless networked systems are in between. A mix often makes sense.

Yes, with modern encryption, but it depends on the security of the device. For sensitive areas a second factor remains sensible.

Yes. Electronic cylinders and fittings can often be retrofitted without new cabling. Check the door's dimensions and fire safety requirements.