What happened
Replaying the persistence-as-a-service intrusion / start not stated; reported May 2025
- 1
Leaked management account key
Date not stated; within one 150-minute windowA leaked long-term access key of an IAM user in the management account is used from five IP addresses within 150 minutes to enumerate Amazon SES settings, attempt an EC2 security group, create IAM users with administrative permissions and make temporary credentials; console sign-ins come from an address in Telegram's network, by an identity the write-up does not name.
- 2
A function that makes IAM users
Date not statedA Lambda function with a new execution role policy and an HTTP API Gateway trigger is created, its code able to create IAM users on demand, a backdoor Datadog says would outlast the key's revocation; which credential created it, and whether its URL was ever called, are not stated.
- 3
Organization integrations turned off
Date not statedFrom a console session, trusted access in AWS Organizations is disabled for the services the write-up lists, six by its count; the change affects new accounts only, Datadog does not know why, and which identity held the session is not stated.
- 4
Identity Center user without MFA
Date not stated; the sign-in came laterIn IAM Identity Center a new user and group get a new permission set, the MFA configuration is changed to allow sign-in without MFA and the Amazon Q Developer session duration is extended to 90 days; later the new user signs in with a password only.
IAM users with administrative permissions, a Lambda function behind an HTTP API able to create IAM users, and an IAM Identity Center user that signed in with a password only after the MFA requirement was turned off; trusted access for organization services disabled. The write-up describes no data access, no email sent and no destruction.
Where the provider's controls fit in a management account
Service control policies do not apply to the management account, so the controls that bind its IAM users are their own. A permissions boundary that accepts calls only from your organization's networks would have refused the leaked key from the attacker's addresses, where your keys are meant for use from known networks. Boundaries that deny IAM user creation and administrative grants, federation token requests, Lambda and API Gateway writes, AWS Organizations writes and IAM Identity Center writes would have capped each backdoor, and one that makes every new role carry an approved boundary without IAM writes would have left the function unable to create users. They bind only the identities that carry them: an IAM administrator's key passes, the write-up does not say which identity made the Lambda, Organizations and Identity Center calls, and an IAM user the attacker created without the boundary passes too. An AWS Sign-In policy on the account would have capped the console sign-ins from Telegram's network, where your console use comes from known networks; AWS added it on June 10, 2026, after the write-up of May 13, 2025. IAM Identity Center MFA settings that prompt on every sign-in and block users with no registered device would have bounded the password-only sign-in only while they stood: the attacker changed the MFA configuration before that sign-in, and a principal allowed to change the settings can do the same. Where the key leaked is not stated, and no cloud control reaches that place.
Could this have been prevented?
In hindsight, not as a whole. 1 control would have refused a step outright, and 8 controls would have limited one. 1 step has no cloud-provider control. The counts below are control assessments, not attack stages: several can address the same step. The same replay with the controls in place before the incident shows where each phase would have been refused or bounded.
- 1
- Prevented
- 8
- Contained
- 1
- Not covered
How it could have been prevented
The same replay with the controls in place. Green marks where a step is refused, blue where it is bounded, and red where no control reaches it. Open a row to read the control, why it acts, and its caveat.
Replaying the persistence-as-a-service intrusion / start not stated; reported May 2025
1 · Leaked management account key
1 Prevented3 Contained1 Not covered6 on other clouds2 · A function that makes IAM users
2 Contained3 on other clouds3 · Organization integrations turned off
1 Contained4 · Identity Center user without MFA
2 Contained6 on other cloudsKey facts
- A long-term access key of an IAM user in the organization's AWS management account was used from five IP addresses within 150 minutes; how the key leaked was not stated
- The key was used to enumerate Amazon SES settings, to attempt to create an EC2 security group and to create several IAM users with administrative permissions, directly or through a group
- The attacker created a Lambda function with a new policy on its execution role and an HTTP API Gateway trigger; Datadog later determined that its code could create IAM users on demand and called it persistence-as-a-service, and whether the URL was ever called was not stated
- Several console sign-in events came in a short period from one address in Telegram's network; Datadog suspected a bot and link previews, and which identity signed in was not stated
- From a console session, the attacker disabled trusted access in AWS Organizations for six services by the write-up's count; Datadog could not determine why
- In IAM Identity Center the attacker created a user and a group with a new permission set, allowed sign-in without MFA and extended the Amazon Q Developer session duration to 90 days; the new user later signed in with a password only
Sources
Every fact in this breakdown comes from the public reporting below. The replay is a model of that reporting, not a reconstruction of the environment it describes.
- 1.Tales from the cloud trenches: The Attacker doth persist too much, methinksDatadog Security Labs · May 13, 2025 · ResearchEntry credential and its account, the 150-minute window and five source addresses, SES enumeration, security group attempt, IAM users with administrative policies, login profiles, federation token, Lambda function behind an HTTP API Gateway, console sign-ins from Telegram address space, disabled trusted access for the Organizations services, Identity Center user, group, permission set, MFA setting and session duration, password-only sign-in
The estate in the replay is illustrative and drawn at the level of attack stages; the victim's real environment is not public. Verdicts describe what a control does when it is in place before the incident. They are not an assessment of the victim's security program.