Acme Protect
Keep AI in bounds, everywhere.
Shield sensitive files and AI conversations, and make sure agents only touch what your company's access rules permit.
More AI means more ways for data to leak
88%
of security leaders report confidential files reaching AI tools. (Industry survey)
2.5x
growth in attacks that use AI to scale phishing. (Threat report)
90,000+
agents projected to run inside large enterprises by 2029. (Analyst forecast)
Companies with a dedicated AI oversight program are
2.6x likelier to catch policy breaches early.(Analyst)
Find and fix data exposure before AI spreads it.
Scan connected sources for confidential material, queue what's found for review, and stop overshared files from appearing in Acme answers.
- Starter rules plus your own policies
- Automatic reviews of agents (beta), skills, and stored memory (coming soon)
- Sort alerts faster with a security chat (beta)
Stop attacks aimed at AI.
Detection models trained on AI-specific attacks flag real incidents and keep false alarms to a minimum.
- Alerts that use the knowledge graph to judge what looks unusual (coming soon)
- Flags 93% of prompt-injection attempts in testing
- Sorts real risk from noise in unstructured files 91% of the time
Keep agents on a short leash
Apply agent rules (coming soon) and check each change an agent wants to make before it happens, so it never drifts past the job it was given.
- Separate rules per agent or per person (coming soon)
- Chosen agents run under their own identity for a clean audit
- Limit what agents can do in a browser (beta)
Certifications and attestations
- SOC 2 Type II
- GDPR
- ISO 27701
- HIPAA
- CCPA
- ISO 27001
Where checks happen
Security checks at every step of a request
Nothing goes unlogged. Each check writes to a tamper-evident audit log in real time, and the events can be forwarded to your SIEM as they occur.
Confirm who is asking through your single sign-on provider before any model runs, for people and agents alike.
Answers follow the sharing settings of each source app, so nobody learns something they couldn't already open.
Label confidential material and redact it before it reaches an answer or gets handed to an agent.
Turn away or reroute prompts on subjects your company has ruled out for any AI use.
Examine prompts and tool calls as they run to block hidden instructions and data theft.
Compare every change an agent wants to make against its assigned purpose before touching a live system.
Rolling out AI with the safeguards switched on
“Vandelay Industries placed Acme on top of the office apps its staff already knew, so the company got AI help everywhere without a single tool migration.”
Research
The GUARD model: a playbook for agent risk
A step-by-step method for scoring agent risk and deciding what data agents may read and what they may change. Developed with several security vendors.
Security and Governance Partner Network
Specialist data protection and AI oversight vendors, integrated with Acme.
- Sentinel
- Vaultline
- Bastion
- Northgate
Found a vulnerability?
Researchers who spot a weakness in any Acme product can earn a reward by reporting it through our disclosure program.