The new intelligence law for the secret services has been proposed in the Netherlands — and Bits of Freedom wouldn’t be Bits of Freedom if we hadn’t taken a closer look at it. It probably comes as no surprise: we have serious concerns about our (digital) human rights. In this article, we’ll outline the key points so you, too, know what to expect.

A big shout out to our volunteers for translating our articles into English! 

AI, AI, AI

One thing that immediately stands out about the proposed law is the extensive scope within which the secret services are permitted to use technology. The law refers to “automation,” which sounds a bit like a shift from paper to digital. That term was chosen deliberately, because it ensures that, in principle, all existing and future technology can be covered by it. This therefore also includes the use of AI and algorithms. This would allow the secret services to analyze so-called bulk datasets — enormous amounts of data on millions of people, the vast majority of which consists of data on individuals who are not under investigation by the intelligence services.

The safeguards proposed to mitigate the risks associated with the use of technology in the work of intelligence services are rather limited. For example, there is only an obligation to assess the reliability of the technology’s application. This means that no assessment is required based on other criteria. It also means that no provisions have been made regarding the use of the technology. In other words: An LLM may be deemed reliable, but you can still ask it questions that return far more data than necessary, meaning its use is not proportionate.

Reports from the regulatory authority show that the secret services are already using AI and automated data analysis without having established sufficient (internal) frameworks and safeguards. The fact that this proposal grants the secret services nearly unlimited authority to use this technology does not contribute to the safe and responsible use of technological resources and increases the risk of misconduct.

Where possible, the 'trawl net' will be widened further

The 'thrawl net authority' provided for in the current law—the 2017 Intelligence and Security Services Act—was the main reason why we all voted down the proposal in a 2017 referendum in the Netherlands. This authority allows the intelligence services to monitor our internet traffic on a larger scale. As you know, it was implemented anyway, but some concessions were made. Those concessions are now being removed.

For example, the criterion of using powers “as specifically as possible” will no longer apply; soon, the entire cable may be tapped on an exploratory basis to see if anything suspicious comes through; even this data may be shared with foreign intelligence services; there will be less of a requirement to clearly specify where the tapping takes place and what criteria are used for filtering; the data may be analyzed using AI; and your streaming data will no longer be safe either.

That covers a lot of ground. But essentially, it means there needs to be less justification for collecting data, that data is collected on a larger scale, that there is less accountability required for analyzing this data, and that such analysis may also be carried out by AI or foreign agencies.

Function creep of authority

It is proposed to expand the objectives for which the intelligence services may exercise the far-reaching powers entrusted to them.

The first new objective is to make it easier or faster to deploy capabilities in the future. This means that agencies are already permitted to hack into infrastructure to establish a foothold in anticipation of situations where valuable intelligence might be gathered in the future—in other words, even without a clear justification at this time.

The second new goal is to develop capabilities. These capabilities can be human — for example, “May we record this conversation for training purposes?”—and they can also be technological, such as for training AI. For instance, intelligence agencies will soon be training their AI using your data as well.

Scaling back the regulator

The regulator is being scaled back quite literally: its capacity will be reduced by as much as 30%, and the number of areas subject to binding review or supervisory authority will be reduced. Restrictive measures are also proposed for the reviews themselves, such as the requirement to justify why intervention is necessary and a limitation on what the regulator may consider in advance when reaching a decision.

There is also the option to appeal the regulator’s decision. In practice, this will give the secret services a second chance to exercise their authority. After all, the minister has already granted permission and is appealing the regulator’s decision regarding that permission. This means that if the regulator disagrees with the minister and issues a “no,” there is an opportunity to appeal. If the regulator agrees with the minister and gives its approval, no appeal will be filed, and the authority will be exercised.

The icing on the cake: emergency state law

“If extraordinary circumstances make it necessary,” various obligations regarding assessment, oversight, and accountability may be suspended. So this cannot be done arbitrarily, but the specific cases in which it is permitted are not defined. It is also unclear how we will emerge from this emergency situation. There is no clear measure in place to ensure that the normal rules will apply again as soon as possible. This would prevent us from getting stuck in a state of emergency.

"Obvious opponents" are declared outlaws

The bill establishes a separate regime for so-called “obvious opponents”—organizations with offensive goals and activities. The designation decision designating an organization as an “obvious opponent” must be reviewed by the regulator for a binding ruling, but the law severely limits the grounds on which the regulator may review this decision. Once an organization has been designated as an obvious opponent, prior authorization is no longer required for a period of one year.

The authority to hack

When it comes to the authority to hack, something very unusual happens. This process is broken down into three steps: gaining access, looking around and securing one’s position—which, as it were, deliberately opens the digital door—and, finally, seizing the data. It is only at that final step that the prior review comes into play. That is, of course, strange. It’s as if the police first break into your home and only then go to get a search warrant based on what they’ve found inside.

This is also a problem because the preliminary review serves a safety function. Currently, prior to a hacking operation, the technical risks must be described in the authorization request. Since the authorization requirement is being eliminated in the new proposal, no one will be reviewing whether it’s actually a good idea to break into a piece of digital infrastructure and leave the door open. Keep in mind that the services are already allowed to hack through third parties. So this isn't just about the devices or infrastructure of so-called targets, but also about the people who interact with them, service providers, or other third parties. In other words, it's about the security of our entire digital infrastructure.

The secret services are everywhere

It is striking how this proposal expands cooperation among government agencies.

There is a list of government agencies from which civil servants can be assigned to perform work for the services. This list includes the Royal Military Police, the Benefits Service, the Tax and Customs Administration, the Immigration and Naturalization Service, the Central Agency for the Reception of Asylum Seekers, and the Correctional Institutions Service. When you’re in contact with these government agencies, you may no longer know whether you’re also in contact with the secret services.

Public-private partnerships may be established to exchange data, jointly analyze data, and provide mutual technical and operational support. Extensive cooperation may take place with foreign companies and foreign intelligence and security services. And support may be sought from Dutch government organizations, such as the police, the armed forces, and the Royal Military Police.

A major risk associated with partnerships is that data may leave the services’ closed confidentiality framework, and that the other parties fall outside the scope of this law. While conditions can be imposed regarding what may or may not be done with the data, it remains to be seen whether those conditions will actually be adhered to. Finally, there appears to be a risk that rules could be circumvented by outsourcing, within a partnership, tasks that the agencies themselves are not permitted to perform to parties not subject to the law.

Conclusion

This proposal allows secret services with fewer accountability requirements to collect, process, and analyze more data using AI. This means our data is more likely to end up on the servers of intelligence agencies, where it is subject to less oversight. For this reason, we believe the proposal is completely out of balance and severely lacks protections for fundamental rights.