What should government say when it has strong intelligence about who was responsible for a hostile act, but cannot prove that conclusion in court, or safely explain how it knows?
That question sits at the heart of modern national security attribution, and recent cases in Britain show why. In March 2024, a warehouse in Leyton, east London, storing supplies for Ukraine was set on fire. In September 2026, armed police stopped three suspicious vehicles heading towards RAF Fairford in Gloucestershire, home to many US service personnel. In November 2026, suspected unauthorised drone activity was reported over RAF Lakenheath in Suffolk, which hosts the largest US fighter operation in Europe. Each of these incidents may point towards a hostile state. Intelligence agencies may hold reporting that the public cannot see, while investigators, journalists and open-source researchers assemble other fragments. Those strands may create a coherent picture, but a strong intelligence assessment is not the same as evidence capable of proving a criminal allegation.
The distinction matters because intelligence and evidence serve different purposes. Intelligence helps government and official agencies understand risk and decide whether to disrupt, sanction or warn. Evidence must withstand legal and disclosure rules and challenge in court. A state may therefore have good reason to believe who was responsible while being unable to prove it publicly. It may also be unable to reveal the strength of its assessment without exposing a sensitive source, collection method or operational capability.
Public attribution in national security involves three distinct claims: what the state is doing, what it assesses and what it can evidence. Each is made to a different standard. Recent cases show why credibility depends on being clear about which claim is being made.
Different products, different claims
Intelligence is not simply collected information. It is produced through processes of evaluation, analysis and interpretation before being disseminated to support decision-making. MI5 describes their assessment as an evolving process where, if they may believe that a particular threat warrants more investigation, it will deploy resources to obtain further intelligence, continually adjusting its assessments in light of updated intelligence or events. However, even confident language based on intelligence assessments retains an element of uncertainty. The UK intelligence community’s probability yardstick places ‘highly likely’ at approximately 80–90 per cent, making clear that even a strong assessment is not a statement of certainty.
Intelligence assessment is also vulnerable to context. Existing assumptions can shape how incomplete or ambiguous reporting is interpreted, while political priorities can influence what is collected, emphasised and presented. Iraq demonstrated the danger, as shown by the Butler Review, which examined weaknesses in the validation and presentation of intelligence, while the Iraq Inquiry, chaired by John Chilcot, concluded that judgements about weapons programmes had been expressed with unjustified certainty. The lesson is not that intelligence should be separated from policy. It is that assessment must retain challenge, caveat and professional independence.
Evidence has challenge and independence built into it to answer a narrower question: what can be established through admissible material in a legal process? Prosecutors must be satisfied that there is sufficient evidence for a realistic prospect of conviction, and a court must be convinced of guilt beyond reasonable doubt before convicting.
The distinction becomes particularly important when disclosure is considered. Intelligence may identify suspects, direct lines of enquiry or inform operational decisions, but once material becomes relevant to a prosecution it may attract disclosure obligations. Sensitive intelligence cannot simply be protected because it is inconvenient to reveal. Its handling must comply with the disclosure regime and, where necessary, courts determine whether sensitive material can properly be withheld. Investigators must therefore consider from the outset not only what intelligence tells them, but the consequences of relying upon it within an evidential case. MI5’s guidance explains that intelligence may be used in evidence, while sensitive material can require public-interest immunity procedures.
None of this suggests that intelligence and evidence compete. They are essential parts of the same national security system. Intelligence may initiate an investigation and support preventative action, whereas investigation may develop material capable of meeting the evidential threshold. Sometimes intelligence itself crosses that divide. But it must make the journey.
What the state may know, prove and say
These standards create three levels of public language. First, authorities may say that they are responding to a threat. This is a protective judgement and does not assign culpability. Second, government may say that it assesses a state to be responsible. That is an intelligence judgement, usually expressed probabilistically. Third, police and prosecutors may say that there is evidence to charge named individuals with specified offences and that they believe the offences are linked to a foreign state. That is an evidential claim which remains an allegation until tested in court.
The levels can overlap but should not be confused in public language and communication strategies. An intelligence assessment that a foreign state directed an operation does not prove that a particular defendant committed an offence. Equally, evidence that an individual committed an offence may not establish who commissioned, financed or directed the activity. The National Security Act 2023 reflects this complexity: where an offence includes the foreign power condition, the prosecution must establish the required connection to a foreign power and the defendant’s relevant knowledge.
There is a further difficulty. Even where intelligence supports a strong assessment, government may be unable to explain that assessment publicly without revealing sensitive sources or capabilities, alerting a target or diminishing the future value of an intelligence technique. Sometimes even carefully qualified attribution can allow an adversary to identify how the state knows. Silence, or cautious language, should not therefore be mistaken for an absence of intelligence, even when it frustrates the media or civilians who have knowledge they feel is compelling to attribution. It may reflect a deliberate decision to protect the very means by which national security is defended.
RAF Fairford and the discipline of an open investigation
The September 2026 incident near RAF Fairford illustrates why an operational response should not be confused with a conclusion about culpability. Following reports of suspicious vehicles close to the airbase, armed officers responded by arresting five men. The scale of the response reflected the potential risk. Counter Terrorism Policing nevertheless asked for space to work through the evidence and said it was examining the full circumstances, knowledge and motivation of those arrested. Surprise was expressed, not least of all by the President of the United States, when those subjects were given bail, with conditions.
The picture developed over the following month. Police confirmed that no improvised explosive devices had been found, although petrol was recovered. They subsequently stated that possible foreign-state involvement formed one of several lines of enquiry. On 27 October, two men were charged under the National Security Act 2023: one with assisting a foreign intelligence service, which police said they believed to be Iran, and the other in relation to a prohibited place. Those allegations remain matters for the courts.
The value of the case study lies in the progression. Risk prompted immediate action. Intelligence allowed the UK Prime Minister, Andy Burnham, to comment on potential attribution, and investigation tested competing explanations. Only later were specific evidential allegations made against specific people, which in turn allowed a more confident assessment of the attribution. Restraint at the outset was not indecision. It protected the integrity of an inquiry whose legal character and attribution changed as the material developed.
The Starmer fires and layered attribution
The fires targeting property associated with the former Prime Minister, Sir Keir Starmer, in May 2025 illustrate a different problem. In June 2026, Roman Lavrynovych and Stanislav Carpiuc were convicted at the Old Bailey of conspiracy to damage property by fire; a third defendant was acquitted. That process established criminal responsibility for the physical acts proved at trial.
Wider reporting explored whether others had commissioned or encouraged the attacks and whether the activity formed part of a broader hostile-state pattern. But these are separate propositions. It may be possible to prove who committed an offence, who commissioned it, assess that the method is consistent with a state’s established behaviour and still lack sufficient evidence to say that the state itself directed the operation.
Proxies are useful precisely because they create this distance. Criminals, intermediaries, online recruiters and low-level agents can place several layers between strategic intent and the person carrying out the act. A 2026 Swedish Defence Research Agency study of 70 espionage convictions across 20 European countries found a wider range of actors than the traditional insider, including non-experts and one-time agents, with recruitment combining conventional methods and social media. Attribution may therefore be compelling to the public narrative on one level while remaining incomplete in reality.
Salisbury: assessment and evidence on different timelines
The 2018 attempted poisoning of former Russian Military Intelligence (GRU, now GU) officer Sergei Skripal and his daughter in Salisbury provides the clearest illustration of the three registers. In March 2018, the UK Government said it was ‘highly likely’ that Russia was responsible for the poisoning of Sergei and Yulia Skripal. At the same time, the Prime Minister stressed that the police investigation should be led by evidence rather than speculation and should be given space to continue.
By September, the investigation had produced sufficient evidence for the Director of Public Prosecutions to bring charges against two Russian nationals. The government attribution and the prosecutorial case were related, but they made different claims and developed on different timelines.
Acting without overstating
It should be taken for granted that with news travelling across the globe faster than ever, the government cannot always wait for a criminal conviction before acting. Protective security, diplomatic measures and sanctions may be necessary while an investigation continues, as seen in the expulsion of Russian intelligence officers shortly after the Salisbury poisoning but prior to charges being secured. Yet naming a state too early carries risk. Early attribution may make it harder to assess later reporting objectively, particularly where analysts or investigators favour information that confirms the initial judgement. It can also blur an assessment of state responsibility with an allegation that named individuals acted criminally on that state’s behalf.
Operational independence is therefore essential. Ministers are entitled to make policy and national security judgements, including on intelligence that could never be used in court. Police and prosecutors must nevertheless retain control of investigative and evidential decisions, test competing explanations and apply the appropriate legal thresholds. A political assessment and a police conclusion may legitimately differ because they answer different questions.
Attribution should be treated as a disciplined process across tactical, operational and strategic levels, with uncertainty made explicit. The operating principle is simple: respond to the consequences immediately; assess responsibility through intelligence; and make criminal allegations only when the evidence supports them. Public language should identify which register is being used: ‘we are responding’, ‘we assess or suspect’, or ‘we believe or can evidence’.
Striking the balance
Intelligence, investigation and prosecution answer different questions and should not be asked to do each other’s work. Fairford shows why an urgent protective response need not amount to attribution. The Starmer fires demonstrate how proof of the physical offence may still leave strategic direction unresolved. Salisbury shows that government assessment and criminal evidence can evolve on related but distinct timelines.
Governments must be able to act on assessments that may never be tested in court, while police and prosecutors must be free to follow the evidence wherever it leads. The discipline lies in being open about which claim is being made. In a contested security environment, credibility depends not on claiming certainty too soon, but on being clear about what we know, what we can prove and, sometimes, what we cannot safely say.
Photo by Jamie Street on Unsplash
Dominic Murphy QPM is the former Commander of SO15, the Counter Terrorism Command within Counter Terrorism Policing. Over a 32-year policing career, he led complex CT and national security investigations, including the Salisbury investigation and has served in senior roles within Counter Terrorism for over 20 years. He now advises boards, insurers and public bodies on crisis leadership, state threats, insider risk, investigations and organisational resilience. His work focuses on translating national security experience into practical decision-making for leaders operating under uncertainty.
