iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
AWS’s 90-day deletion rule applies to a specific opt-in project spend limit: when a project reaches its limit, AWS pauses it and stops its resources while preserving its data. If the project owner takes no action within 90 days after that pause, AWS says it permanently deletes the project data. This is not an automatic 90-day deletion policy for every AWS account, and AWS Budgets does not trigger it.
What happens when an AWS project reaches its spend limit?
AWS describes project spend limits as an opt-in control in AWS Settings. An administrator can choose a recommended limit or set a custom one. The limit takes effect immediately, and the setup also offers optional early controls: “Stop new resource launches,” “Pause idle resources,” and “Pause top cost drivers.” These options do not establish that every possible AWS charge is covered.
When the project reaches its limit, AWS pauses the project and stops all its resources, but preserves the project data at that point. To reactivate the project, the owner raises the spend limit; AWS notes that some resources may need to be restarted manually. See AWS’s project spend limits documentation for the current flow and feature details.
When does the 90-day deletion period begin?
The period begins when the project is paused for reaching its spend limit—not when a budget warning is sent. AWS’s account-management documentation states: “If you take no action within 90 days of your project being paused, AWS permanently deletes your project data.” The consequence is conditional on inaction after the pause; it is not a general retention rule for all data in an AWS account.
#1 Best Overall
AWS’s published description does not spell out every resource type or recovery edge case. Do not assume that every resource can be restored in exactly the same state, or that raising the limit will restart everything automatically.
How the AWS cost controls differ
| Control | Scope and function | Timing and limitation |
|---|---|---|
| Project spend limit | An opt-in control for an AWS project; pauses the project and stops its resources at the limit, with optional early controls. | Takes effect immediately. No action within 90 days of the pause leads to permanent deletion of project data, according to AWS. |
| AWS Budgets | Tracks configured costs or usage and can send notifications or take configured actions. | Information updates up to three times a day, typically 8–12 hours after the previous update. Charges can accrue before an alert arrives. |
| Agent-level budget policy and cutoffs | Controls a designed agent workflow through pre-invocation checks, session cutoffs, tool-call limits, throttling, or memory guardrails. | Must be implemented and monitored; AWS does not say these controls are automatic for every agent. |
| Cost anomaly history | Helps investigate detected irregular cost patterns. | Anomaly records are available for up to 90 days and archive after that period. This is separate from project-data deletion. |
Can AWS Budgets stop an AI agent from overspending?
AWS Budgets is useful for account oversight and escalation, but it should not be treated as an instantaneous per-agent spending meter or a guaranteed stop at an exact dollar amount. AWS says budget information updates up to three times daily, typically 8–12 hours after the previous update. It also warns that billing and usage delays can mean more charges accrue before a notification arrives, and costs may continue changing after notification. Budgets can be configured with actions—for example, AWS documents applying a custom IAM policy—but that does not remove the timing caveat. See AWS Budgets documentation.
Rank #2
For an agent that can launch resources, call paid services, or trigger work in other systems, put preventive controls in the workload path as well as monitoring. AWS’s Agentic AI Lens identifies cost risks such as reasoning-loop token spikes, bursts of tool invocations, and memory growth. Its guidance is to establish limits before invocation and enforce them outside the agent’s own control loop.
The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →How to set spending controls for an AI agent on AWS
Use layered limits rather than relying on one account-level alarm. The exact thresholds depend on the workload, its services, and the cost of each operation; AWS’s guidance does not prescribe universal dollar amounts.
Rank #3
- Set limits at multiple horizons. Check budget before each cycle, task, and day. A task-level ceiling can contain one runaway job without consuming the entire daily allowance.
- Bound each session. Cut off a session when it reaches a maximum iteration count or cumulative token cost. This limits repeated reasoning loops that otherwise continue until a task finishes or an external failure occurs.
- Cap tool calls. Set invocation limits for tools and consider their downstream costs: a call may incur its own service charge and create results that require additional model tokens to process.
- Constrain memory growth. Define a context or memory ceiling and trigger summarization or pruning before accumulated state drives token use higher.
- Throttle before the hard stop. Reduce request rates as usage approaches a daily limit, then apply a hard cutoff for sessions that continue consuming resources.
- Monitor the controls themselves. Track budget utilization, cutoff events, and throttling events so operators can distinguish normal limits from a stuck workflow or misconfigured policy.
AWS describes Amazon Bedrock AgentCore Policy as a way to apply Cedar policies at an AgentCore Gateway boundary, helping prevent an agent from bypassing policy through prompt manipulation. The same AWS guidance discusses Amazon Bedrock Guardrails for topic restrictions. These are implementation approaches, not an automatic hard spending cap for every agent. Confirm service behavior and availability for the relevant account and Region before relying on a particular control.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Why two different AWS features mention 90 days
The project deletion period and cost-anomaly history are unrelated windows with different objects and outcomes. The project rule concerns data for a project paused after reaching its spend limit. Separately, the Cost Explorer GetAnomalies API reference says detected anomalies are available for up to 90 days; AWS’s anomaly investigation guide says anomaly data archives after 90 days. Anomaly retention does not trigger project deletion or govern project recovery.
Quick Recap
Best Value
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
