
Who Is Behind the Attacks? Attribution Explained
Every incident in this hub carries an Attribution line. This page explains what those lines mean, who the named groups are, and why so many attacks have no attribution at all. Attribution is not a single thing. It ranges from a court finding, through a government statement based on intelligence, through a threat-intelligence firm's medium-confidence assessment, to a claim on Telegram by a group that may not have done anything. Treating those as equal is the most common error in cyber reporting. This page keeps them apart — and names the groups that do appear in the verified record, with the evidence for each.
In short
Attribution in this hub has four levels. Official: a government authority or agency has named an actor — Shahid Kaveh for the October 2026 France fridge attack (French Foreign Minister, 9 November 2026); Russian / Russia-linked hackers for the Norway dam (PST chief, 13-14 August 2025); Iranian-affiliated actors for the US PLC campaign in general (CISA advisory AA26-097A). Reported: reputable media have reported a link no authority has confirmed — the Iran link for the UK power generator (Telegraph / Financial Times), Scattered Spider for M&S and the Co-op. Claimed: a group says it did something, with no independent verification — Qilin for Asahi, Embargo for ChipSoft, Interlock for DaVita, NoName057(16) / Z-Pentest for the November fridge wave, 'Scattered Lapsus$ Hunters' for Jaguar Land Rover (not regarded by investigators as the perpetrator). None: no actor named — Ahold Delhaize, the November fridge wave, the EFS blood bank, the UK power generator (officially), Boston Scientific. Conflicting: two credible sources name different actors — Poland, where CERT Polska says Berserk Bear / FSB and ESET / Dragos say Sandworm / GRU (medium confidence). No attribution in this hub is stronger than its evidence, and no claim is presented as a finding.
What attribution means — and what it does not
Attribution: This page is about the concept; each incident's attribution is on its own page. Attribution is the process of deciding who carried out a cyberattack, and it is harder than it looks. Attackers hide behind rented infrastructure, stolen credentials, copied tools and false flags. A single attack can involve an initial-access broker who sold the entry, a ransomware affiliate who deployed the malware, and a group that laundered the payment — three different actors, one incident. Attribution is therefore not a single answer but a judgement, made on evidence of varying strength, by different kinds of authority, for different purposes.
This hub uses four levels, and the difference between them is the point. Official means a government authority or agency has named an actor on the record. Reported means reputable media have reported a link that no authority has confirmed. Claimed means a group has said it did something, with no independent verification. None means nobody has been named. There is also a fifth situation — conflicting — where two credible sources name different actors, as in Poland. And there are two qualifiers that matter: confidence (ESET's 'medium confidence' for Sandworm is a formal statement about how sure the analysts are) and scope (CISA's attribution of a campaign in general is not an attribution of a specific incident). A reader who understands those distinctions will not be misled by the next claim they see. A reader who does not will be misled by almost every claim they see.
The state-linked groups in the verified record
Attribution: See each group below. Four state-linked groups appear in the verified record of this hub. Sandworm (GRU Unit 74455 / APT44) is Russia's military intelligence cyber unit, known for the first cyber-driven power blackouts in Ukraine — 23 December 2015 (about 225,000 customers, BlackEnergy / KillDisk) and 17 December 2016 (Kyiv transmission substation, Industroyer / CRASHOVERRIDE) — and for NotPetya in 2017, which caused about $10 billion in global damage. ESET and Dragos attribute the Poland December 2025 wiper attack to Sandworm with medium confidence. ESET also assessed that UAC-0099 (Earth Sirrush) handed off to Sandworm in the Ukraine grain-logistics wiper attack of 17 September 2025.
Shahid Kaveh is an offensive cyber unit inside the Electronic and Cyber Centre of Iran's Islamic Revolutionary Guard Corps. France attributed the October 2026 fridge attack to Shahid Kaveh on 9 November 2026, based on infrastructure characteristics, signature tools, a propaganda outlet linked to the unit, and Persian / Farsi traces. Three alleged associates were named in reporting: Hamid Reza Lashgarian, Mahdi Lashgarian and Reza Salarvand. Berserk Bear (Static Tundra) is assessed by CERT Polska as linked to Russia's FSB (Center 16), and CERT Polska attributes the Poland attack to it. The IRGC Cyber-Electronic Command / CyberAv3ngers ecosystem is named in CISA advisory AA26-097A as the Iranian-affiliated actor behind the US PLC campaign in general. CyberAv3ngers is known for earlier attacks on Israeli and American water systems. These are the only state-linked groups in the verified record. Any other name a reader sees attached to these incidents is either reported, claimed or invented.
The criminal groups: ransomware as a business
Attribution: Claimed or reported — criminal groups are not state-attributed. The criminal groups in this hub are ransomware operations, and they work differently from state groups. They want money, not sabotage, and their business model is ransomware-as-a-service: a core team builds the malware and takes a cut, while affiliates break in, deploy it and negotiate payment. ALPHV / BlackCat hit Change Healthcare in February 2024 and exit-scammed in March 2024 after a reported $22 million payment. DragonForce was the ransomware deployed against Marks & Spencer on 24 April 2025, used by affiliates linked by researchers and media to Scattered Spider (UNC3944), a group known for social-engineering help desks. Interlock hit DaVita in March-April 2025 and claimed more than 20 TB exfiltrated. Qilin, active since 2022 and described as Russia-linked in secondary reporting, hit Asahi and was the most active group in the food and agriculture sector in 2025. Embargo claimed the ChipSoft hospital-software attack in April 2026, with about 100 GB claimed stolen. REvil hit JBS in 2021 and was attributed by the FBI.
The criminal groups are, in one sense, easier to attribute than the state groups: they claim their attacks publicly, on leak sites, because publicity is part of the extortion. But a claim is still a claim. The group that claims an attack is not necessarily the group that carried it out — affiliates, imitators and opportunists all claim attacks — and the claim tells the reader nothing about who the affiliates were or who else was involved. The JLR case is the warning: a group calling itself 'Scattered Lapsus$ Hunters' claimed the attack on Telegram, investigators do not regard that group as the actual perpetrator, and a New York Times investigation pointed to Russian hackers instead. A Telegram claim is the weakest form of attribution in this hub, and it is labelled as such wherever it appears.
Why so many attacks have no attribution
Attribution: None — for the incidents listed below. Five incidents in this hub have no attribution at all: Ahold Delhaize (no actor, no attack type disclosed), the November 2026 fridge wave (no authority has named anyone), the EFS blood bank (no claim, no attribution), the UK power generator (officially unattributed, with only media reporting of an Iran link), and Boston Scientific (no group claimed it). There are good reasons for that. Some attacks leave little forensic evidence — a weak password and a defrost command do not leave the kind of infrastructure trail that identifies an actor. Some investigations are still running, and authorities will not name an actor until the evidence is ready, as France showed by waiting a month before attributing the October attack. Some attacks are not worth attributing publicly, because the victim is a small generator or a single hospital vendor and the diplomatic cost of naming a state outweighs the benefit.
There is also a structural reason: the actors who benefit from ambiguity cultivate it. A wiper that destroys data leaves less evidence than ransomware that negotiates. A hacktivist claim that is deleted within hours leaves less than a leak-site listing that stays up. And a method that is published — exposed controllers, default credentials, defrost commands — can be copied by anyone, which makes the next attack harder to attribute than the first. The November wave may never be attributed, as Belgium's Centre for Cybersecurity said. That is not a failure of investigation. It is a property of the attack: when the method is public and the entry point is a default password, the list of possible attackers is everyone.
Medium confidence, and why it matters
Attribution: ESET / Dragos on Poland — medium confidence. Confidence levels are the most misunderstood part of attribution. When ESET and Dragos attribute the Poland attack to Sandworm with medium confidence, they are making a formal statement: the evidence points to Sandworm, but it is not conclusive. Medium confidence is not a hedge or a euphemism. It is a defined level in intelligence assessment, meaning the evidence is credible and plausible but could be explained another way — for example, by shared tooling, by a false flag, or by a different group using Sandworm's tools. CERT Polska's attribution to Berserk Bear / FSB is stated without a published confidence level, which does not make it stronger or weaker — it makes it differently expressed.
The Poland conflict is the case study for why confidence matters. Two credible sources, two different Russian services, one attack. If readers treat both attributions as certainties, they contradict each other and one must be wrong. If readers treat them as assessments with different evidence and different confidence, they can both be understood: CERT Polska investigated the incident and assessed FSB linkage; ESET and Dragos analysed the malware and assessed Sandworm linkage with medium confidence. The two may reflect different parts of a multi-target operation, different infrastructure, or genuinely different judgements about the same evidence. Until the accounts are reconciled, the honest statement is both, with their sources and their confidence. That is what the Poland page does, and it is what this page recommends as the standard for every contested attribution.
How to read the next claim you see
Attribution: Guidance, not a finding. The next claim about a cyberattack will arrive soon — on social media, on a leak site, on Telegram, in a news report. This page offers a short test for it. First, who is making the claim, and what is their interest? A group claiming its own attack has an interest in publicity. A government attributing an attack has an interest in deterrence and diplomacy. A vendor selling a product has an interest in a scary story. Second, what is the evidence? Infrastructure, tooling, code, language traces and a propaganda outlet — the elements in the ANSSI report on Shahid Kaveh — are evidence. A Telegram post repeating details that were already public is not. Third, what is the confidence and the scope? Medium confidence about a campaign in general is not high confidence about a specific incident. Fourth, has anyone else confirmed it? An attribution that stands alone, contradicted or unconfirmed by every other authority, should be treated as provisional.
The final rule is the simplest: if a claim would be convenient — if it confirms what you already believe about who the enemy is — be more sceptical, not less. The October 2026 fridge attribution was inconvenient in one sense and convenient in another: it named Iran, which fitted a pattern, but it took a month, rested on technical evidence, and was limited to one attack in one country. The claims that should worry readers are the ones that are instant, total and convenient: every incident, one campaign, one enemy, no evidence. There is no such campaign in the verified record of this hub. There are separate attacks, by separate actors, with separate motives, in the same two years — and that, unglamorous as it is, is the truth.
Frequently asked questions
A government authority or agency has named an actor on the record — for example, France attributing the October 2026 fridge attack to Shahid Kaveh, or Norway's PST blaming Russian hackers for the dam attack.
Official is a government or agency naming an actor. Reported is reputable media reporting a link no authority has confirmed. Claimed is a group saying it did something, with no independent verification. Official is strongest; claimed is weakest.
GRU Unit 74455 (APT44), Russia's military intelligence cyber unit. It caused the 2015 and 2016 Ukraine power blackouts and NotPetya in 2017. ESET and Dragos attribute the Poland December 2025 wiper attack to Sandworm with medium confidence.
An offensive cyber unit inside the Electronic and Cyber Centre of Iran's Revolutionary Guard. France attributed the October 2026 supermarket fridge attack to Shahid Kaveh on 9 November 2026.
UNC3944, a group known for social-engineering help desks to reset passwords. Researchers and media linked it to the Marks & Spencer and Co-op attacks using DragonForce ransomware. The link is reported, not official.
CERT Polska attributes it to Berserk Bear (assessed as FSB-linked); ESET and Dragos attribute it to Sandworm (GRU) with medium confidence. The two name different Russian services. Both are credible sources, and the conflict is presented honestly rather than resolved by picking one.
Some leave little forensic evidence, some investigations are still running, some are not worth attributing publicly, and some methods — like a default password and a defrost command — can be copied by anyone, making the next attack harder to attribute than the first.
Ask who is claiming it and why, what the evidence is, what the confidence and scope are, and whether anyone else has confirmed it. Be most sceptical of claims that are instant, total and convenient — every incident, one campaign, one enemy, no evidence.
Figures are given with their snapshot time and source. Developing items are marked as developing. Last checked Tuesday 17 November 2026, about 12:50 PKT.