Who did it? How attribution really works
Attribution and confidence · facts checked 2026-10-03
The facts
- The US intelligence community's analytic standards, ICD 203, require assessments to state the analysts' confidence. (ODNI)
- MITRE ATT&CK lists the other names each threat group is tracked under by different security companies. (MITRE)
- MITRE ATT&CK tracks more than one hundred threat groups. (MITRE)
- In 2023 Microsoft moved to a weather-themed naming system for the threat actors it tracks. (Microsoft)
The panel
Jordan Pike AI host
ODNI says the US intelligence community's analytic standards, ICD 203, require assessments to state the analysts' confidence. MITRE says its ATT&CK knowledge base tracks more than one hundred threat groups, and it lists the other names each group is tracked under by different security companies.
Jordan Pike AI host
Microsoft says that in 2023 it moved to a weather-themed naming system for the threat actors it tracks.
Jordan Pike AI host
So the headline says a particular group did it. Rina, how much weight does that sentence deserve before anyone does anything on Monday?
Rina Takahashi AI analyst
Treat it as someone's claim, every time. Ask who is making it, what evidence they are leaning on, and how sure they say they are. That is why intelligence standards demand a stated confidence level. Attribution usually rests on patterns like shared infrastructure, tooling habits and timing, and each of those can be shared, borrowed or copied. A confident headline with no confidence statement behind it is just a guess in a nice suit.
Dex Moreno AI analyst
And from the attacker's seat, that's an opening. If attribution leans on habits, an intruder could deliberately reuse another crew's habits and leave investigators arguing about the wrong name. Muddying the trail like that is, I'll admit, an eight out of ten for elegance. The victim deserved better, obviously.
Jordan Pike AI host
Dex, we are not rating it out of ten.
Rina Takahashi AI analyst
Then there's the naming mess. Different security companies track the same activity under different names, which is why MITRE keeps that alias list. Add a weather-themed scheme and you've got one more vocabulary on the pile. The industry could have agreed on one name per group. Instead we got a translation problem and a forecast. So before you assume two reports describe two groups, check whether they're the same group wearing different labels.
Dex Moreno AI analyst
Defenders can cut the path off without waiting for a name. An attacker needs a way in, a foothold, and then somewhere to move to, like stored secrets or trusted accounts. Those steps look much the same whoever is behind them. So build detection around behaviour, and treat the group name as context, not a trigger. Whatever the label, close the way in and go hunting for a foothold that may already exist.
Rina Takahashi AI analyst
My action: when a report names a culprit, write down who attributes it and at what confidence, and check the alias list before you merge or split anything in your own tracking. And when your board asks who did it, the honest answer is a claim, a confidence level, and what you're doing either way.
Jordan Pike AI host
So: Monday morning? Read the attribution as a claim, check the aliases, and defend against the behaviour rather than the name.





