Imagine the U.S. government contracted with a private company to assemble a small team, fly it overseas, break into an office, and wreck enough of the operation to shut it down for months, perhaps for good. Washington would approve the target, lawyers would establish the limits, and other federal agencies would be consulted before the team went in.
That is, more or less, what the Trump administration has authorized in cyberspace. President Donald Trump last week approved a policy allowing selected American companies to conduct cyberoperations against foreign criminal organizations on the government’s behalf. Until now, the United States generally kept offensive cyber operations of this kind within government. Under the new policy, contractors could penetrate foreign systems for intelligence and, with approval from the Justice and Homeland Security Departments, disrupt, manipulate, or destroy them.
The administration’s argument is straightforward: Cybercrime is growing faster than the government can reasonably attack every threat itself, but American companies employ formidable cyber talent. Why not tap into that capability?
It is a fair question. But it addresses whether private companies can do this, not whether the government should make private offensive cyberoperations a standing instrument of national policy.
The administration has not created a free-for-all. The memorandum promises the federal government “complete oversight and control.” Companies will be vetted, operations will require written approval; operators must post a $1 million bond; and actions expected to cause death, serious injury, or effects equivalent to an armed attack will be prohibited. A classified process is intended to prevent contractors from interfering with operations already being conducted by the military, intelligence agencies, or law enforcement.
Those safeguards matter. The military and national security agencies employ contractors, including for sensitive work, and contractors can perform difficult missions effectively only when responsibilities are clear and oversight is strong. Recent events also make it difficult to argue that this administration simply cannot manage complicated operations. Putting aside the wisdom, legality, or longer-term consequences of the operations themselves, the capture of former Venezuelan president Nicolás Maduro involved months of intelligence collection, rehearsals, and coordination among multiple agencies. The opening strikes in the Iran war, which killed Ayatollah Ali Khamenei and much of Iran’s senior leadership, were likewise contingent on intelligence, deception, and precision targeting working together.
But what happens when something designed for exceptional cases becomes routine? Will these safeguards remain rigorous after the 50th operation?
The term “criminal organization” introduces another problem: Who or what will qualify as a criminal organization? The Trump administration’s 2026 Counterterrorism Strategy, released in May, blurred the line between demonstrated movement toward violence and broader ideological or political categories. That does not mean the same thing will happen here. But once private companies are authorized to conduct offensive operations, the criteria used to define an eligible target become critical.
Even after a group is properly identified as criminal, determining whether it operates independently or has ties to a foreign government can be difficult.
Cyber operations are not always easy to contain. Stuxnet, the U.S.-Israeli cyberweapon reportedly ordered during the George W. Bush administration and expanded under President Barack Obama, was designed in the late 2000s to sabotage centrifuges at Iran’s Natanz enrichment facility. Yet the malware escaped the plant, spread beyond Iran, and was eventually detected on tens of thousands of computers worldwide. Not every offensive operation will become another Stuxnet, but the caution is worth noting: Even sophisticated operators cannot always predict where malicious code will travel or how it will behave once released.
Artificial intelligence adds another oversight problem. Trump’s policy does not authorize autonomous AI agents to conduct these operations, but private-sector cybersecurity is becoming increasingly automated. Earlier this month, Britain’s AI Security Institute reported that during testing an advanced Anthropic model tried to insert malicious code into a real open-source project, created false identities when it encountered resistance, and altered records of its earlier actions. In the same evaluation, an OpenAI model also took unsanctioned actions outside the intended test environment. The companies stressed that safety controls had deliberately been reduced and no one was harmed, but the episode illustrates that government approval before an operation starts is not necessarily the same thing as control over every action taken at machine speed once it is underway.
There is also a policy contradiction worth confronting. Since Trump returned to office, the Cybersecurity and Infrastructure Security Agency (CISA) — the government’s principal civilian cybersecurity agency and its main bridge to critical infrastructure — has lost roughly a third of its workforce, and other parts of the federal cyber apparatus have experienced personnel cuts and leadership disruption. CISA does not conduct the offensive operations, so its contraction does not directly explain a shortage of offensive capacity. But the sequence still matters, given that the administration is arguing that the threat has become too large for the government to handle.
Perhaps that is intentional. Trump officials may believe government should set objectives and control operations while purchasing more capability from outside. That model may work in isolated cases. The danger is what happens when repetition, speed, and scale begin to test the safeguards designed to control it.
Brian O’Neill, a retired senior executive from the CIA and National Counterterrorism Center, is an instructor on strategic intelligence at Georgia Tech. His Safehouse Briefing Substack looks at what’s ahead in global security, geopolitics, and national strategy.
Who gets to decide what crime is being committed and who's the offender? While I have no problem with American hackers doing some hacking against bad foreign actors, I can see this spinning out of control.
Since the American cyberoperators will be aligned with Trump, it is a very real possibility that they go after legitimate operations because they're Trump's opposition. He has already weaponized the entire US government against his domestic enemies, this gives him another weapon to take it internationally.
My greatest fear about this is eventually the government will go after private U.S. citizens or legitimate American businesses just because they don’t like something or have an ax to grind with someone.