How is Emergency Freeze different from a block-mode budget?

A block-mode budget stops calls once a specific limit is hit. Emergency Freeze stops everything, immediately, with no limit to reach first — it's for 'stop right now while I investigate,' not for ongoing enforcement.

How fast does a freeze take effect?

Within about 10 seconds. The freeze check runs inline in the same cached auth lookup as your plan and rate limit, which refreshes every 10 seconds — the same staleness window that check already has for other fields.

Does freezing my account affect my dashboards or data?

No. Freeze only blocks new proxied calls going forward. Your historical data, dashboards, budgets, and settings are all unaffected and remain fully visible.

What triggers an automatic freeze?

Only if you've opted in via 'Auto-freeze on high-risk signals' in Settings. When on, a spend-velocity spike, a high-severity compromise-risk signal, or an unexplained provider billing gap freezes the account automatically instead of only alerting. Off by default.

Can I still unfreeze if I'm the one who caused the freeze by mistake?

Yes. Unfreezing is self-service from Settings → Emergency Freeze, same as freezing — there's no support ticket or waiting period.

Didn't Google/OpenAI/Anthropic add hard spend caps of their own?

Yes, since April-July 2026 all three now offer some form of account- or workspace-wide hard cap. They're real progress, and also genuinely coarse: whole-account (not per-feature or per-team), single-provider, and — by the providers' own documentation — 10 to 20 percent of spend can still land in the window before enforcement catches up. Emergency Freeze doesn't replace those; it's the granular, cross-provider, atomic-reservation layer on top.

How do I hard-stop all AI spend immediately?

Emergency Freeze blocks every proxied call for your account instantly, regardless of budgets, plan, or enforcement mode — a manual or automatic kill switch for when you need to stop spend right now.

Emergency Freeze is a hard stop: every proxied call for your account is rejected with HTTP 403 until you unfreeze, regardless of what any budget, plan, or enforcement mode says. It exists for the moment a budget isn't the right tool — you suspect a compromised key, a runaway agent, or something else you need to stop right now while you figure out what happened.

This was built for a failure mode that keeps recurring: a compromised cloud credential gets used to run image or video generation at scale, and the bill — sometimes five or six figures — only surfaces after the damage is done, with no cap that could have stopped it in advance. Emergency Freeze is Cognocient's answer for spend routed through its proxy: one click stops everything, no limit has to be reached first, and — unlike a setting on the compromised account itself — it lives behind Cognocient's own login, outside the blast radius of a stolen proxy key.

What's changed on the provider side, and what still hasn't

Between April and July 2026, Google, OpenAI, and Anthropic each shipped some form of their own hard spend cap — a real, overdue fix, and worth being accurate about rather than pretending it didn't happen:

Provider	What shipped	Scope	Enforcement lag
Google Gemini	Mandatory, non-disableable billing-account cap (Tier 1 $250/mo, Tier 2 $2,000/mo, Tier 3 $20K–100K+)	Whole billing account, all projects under it	~10 minutes of overage before it bites
OpenAI	Optional hard limit at org/project level	One org or project	10–20% of spend can leak past the cap (async billing)
Anthropic	Workspace-level hard cap	One workspace	Not instant — same class of async-billing lag

Coarse, single-provider, still not instant

None of these give you a per-feature or per-team ceiling, none of them work across providers (a Gemini cap does nothing for your OpenAI spend), and hitting one stops everything under that account — there's no graceful degrade, just a full stop for every legitimate workload sharing it. They're a real backstop. They're not what Emergency Freeze + budget enforcement + dollar-denominated velocity give you: atomic reservation before the call, scoped to a feature/department/key, across every provider you use, with a manual override that works even if a provider's own cap setting is what gets misconfigured or reset.

Freezing your account

Go to Settings → Emergency Freeze and click Freeze all spend now. You can optionally leave yourself a reason (for your own audit trail — it's not sent anywhere). The freeze takes effect within about 10 seconds across every process serving your traffic.

Every call made while frozen gets:

{
  "error": {
    "message": "This account's spend has been frozen.",
    "type": "account_frozen"
  }
}

A red banner showing Spend is frozen appears across every page of the dashboard while the freeze is active, with a one-click Unfreeze button right there — you don't have to go back to Settings to clear it.

Unfreezing

Click Unfreeze spend from the same Settings panel, or the Unfreeze button on the banner. This is self-service in both directions — freezing is not a support-mediated action, and neither is clearing it.

Automatic freeze

By default, Cognocient only alerts on a spend-velocity spike, a compromise-risk signal, or a shadow-spend reconciliation gap — it never freezes your account on its own. If you'd rather those situations stop spend automatically instead of just paging you, turn on Auto-freeze on high-risk signals in the same Settings panel.

Auto-freeze is off by default deliberately — an automatic hard stop should never surprise an account that never asked for one. Turn it on once you've seen a few real alerts and are comfortable with what triggers them.

When auto-freeze applies, it can be triggered by:

Trigger	Source
Spend-velocity spike	Dollar velocity breaker — 10× your normal $/min baseline
High-severity compromise signal	Compromise Risk — e.g. a key's first-ever image-generation call
Unexplained provider billing gap	Shadow Spend Reconciliation

Every freeze — manual or automatic — is logged with who or what triggered it and when, visible in your freeze history.

What Emergency Freeze does not do

Scope: proxy-routed spend only

Emergency Freeze blocks calls that go through Cognocient's proxy. It has no effect on a raw provider API key called directly, or on a compromised provider account acting outside Cognocient entirely — see Shadow Spend Reconciliation for how Cognocient tries to catch that separately, by comparing what your provider billed against what the proxy actually saw.