Summary
- Germany’s cabinet has approved legislation updating BND and BfV powers for AI analysis and cyber operations.
- Strategic intelligence data could be retained for longer, while narrowly defined active measures could be used against imminent cyber threats.
- The additional powers would sit alongside expanded prior oversight by the Independent Control Council.
Germany is preparing to write artificial intelligence and active cyber intervention directly into federal intelligence law, expanding the technological powers available to its security agencies while placing the new capabilities inside a more formal system of judicial and independent oversight.
The German federal government approved the reform package on 12 August, covering both the Federal Intelligence Service, or BND, and the domestic Federal Office for the Protection of the Constitution, known as the BfV. Among the changes, the legislation would explicitly regulate the use of AI applications to analyse intelligence data and allow the BND to retain telecommunications content for up to six months and traffic data, including connection records, for as long as 12 months during strategic intelligence work.
Federal intelligence agencies would also gain authority for tightly limited active protective measures when another body, such as the police, cannot prevent a threat with comparable effectiveness. In its explanation of the bill, the government gives the example of disabling a foreign server from which an imminent cyberattack is being launched, moving the legal framework beyond information gathering and towards direct disruption of hostile infrastructure.
Those operational powers are being introduced alongside tighter external controls because previous Constitutional Court judgments have forced Berlin to define surveillance authority more precisely. The Independent Control Council would take over the responsibilities of the existing G10 Commission and gain jurisdiction across the federal intelligence services, while particularly intrusive measures would require prior approval.
AI changes what existing data can reveal
Although intelligence agencies have used automated analysis for years, generative systems and newer machine learning tools can substantially increase what can be extracted from information that authorities already possess. Large collections of communications data can be searched, correlated, translated, classified, or connected more quickly, which means the practical reach of a lawful dataset can change even when the statutory collection power itself does not.
That creates a harder proportionality question than simply deciding whether an AI tool is permitted. An agency that once needed substantial analyst time to identify connections across many records may be able to automate much of that work, while systems capable of entity resolution and pattern detection can reveal relationships that would previously have remained buried in volume. Rules governing collection therefore have to operate alongside controls over analysis, retention, access, and the decisions made from resulting intelligence.
Germany’s proposal reflects a wider shift across European government, where AI is moving into policing, defence, fraud detection, border management, tax administration, and other functions with substantial consequences for individuals. Much of the EU regulatory debate has centred on private providers and commercial deployments, but national security and intelligence sit inside a more complicated mixture of national law, constitutional protections, judicial review, and exemptions from parts of the EU AI framework.
The reform also extends that debate beyond model governance because the proposed cyber powers would allow intelligence work to affect third party infrastructure directly. Modern attacks routinely pass through compromised servers, cloud accounts, rented infrastructure, and devices belonging to organisations that may have no connection with the attacker, so an apparently simple action against a hostile server can raise difficult questions around attribution, jurisdiction, and collateral disruption.
Cyber defence moves closer to hostile infrastructure
European security policy has traditionally concentrated heavily on hardening domestic networks, sharing threat information, and improving incident response, although governments have also developed offensive and intelligence capabilities largely outside public view. Germany’s legislation would put a more explicit legal basis around some intervention at the point where an external system is being used to create an imminent threat.
Cloud operators, communications providers, and cyber security companies could consequently find themselves closer to state operational activity, particularly where infrastructure under their control is compromised or becomes part of an attack chain. Clear procedures for evidence, notification, escalation, and cooperation will determine whether additional state capability improves resilience without creating unnecessary uncertainty for infrastructure providers.
Longer retention periods create a related governance burden because the usefulness of historic data depends on the controls surrounding it. Six or 12 months of strategic intelligence can help investigators reconstruct networks after an event, yet retaining more material also increases the amount of sensitive information that has to be protected, audited, and eventually deleted under enforceable rules.
Parliament will now determine the final shape of the legislation, while its durability will also depend on whether the new provisions satisfy Germany’s constitutional standards once they are applied. The cabinet proposal nevertheless puts AI analysis and active cyber operations into the formal architecture of intelligence work, rather than leaving them to develop as technical capabilities around laws written for an earlier communications environment.
