Keep private data private.
Catch sensitive content before it reaches a model.
Browser security ↗ LLM Gateway ↗AI SECURITY. BEFORE THE ACTION.
Hawk checks AI requests and tool calls against your rules. Block sensitive data, stop unauthorized actions, and record the outcome.
For the prompts your people send.
And the actions your agents take.
Stop sensitive data before a public AI chat is submitted.
↘02 / LLM GatewayWhat leaves for a model.Keep sensitive data out of AI requests.
↘03 / MCP GatewayWhat a tool can do.Stop actions beyond an agent’s permissions.
↘04 / GitHub controlsWhere agents touch code.Set boundaries for repository access and changes.
↘HOW HAWK WORKS
Connect your AI workflows to Hawk. It checks each routed request before it reaches a model, calls a tool, or changes a repository.
Catch sensitive content before it reaches a model.
Browser security ↗ LLM Gateway ↗Check the tool and action against the agent’s permissions.
See the MCP Gateway example ↗Decide which repositories agents can read or change.
See the GitHub example ↗ONE HAWK. FOUR PROTECTION POINTS.
Reduce AI risk across browsers, models, tools, and code. Choose a surface to see what Hawk checks and blocks.
01 / Browser security
Someone copies customer data into ChatGPT or another public AI chat. Hawk’s browser extension checks the draft and blocks submission before that data is sent.
Prompts, pasted text and supported uploads in managed AI browser sessions.
Enforced on supported AI sites where the Hawk extension is installed, enabled and managed.
Explore browser protection ↗Summarize these accounts for my meeting:
This draft has not been submitted.
Not sent to ChatGPTHere’s your summary.
Your team has three product updates to share.
02 / LLM Gateway
Your internal AI app sends a request to a model. Hawk’s LLM Gateway catches customer account numbers and blocks that API request before it reaches the provider.
Customer information, credentials, confidential prompts and model interactions.
The customer file is attached. Hawk checks this request before it leaves.
Here’s your summary.
Three product updates are ready to share with the team.
03 / MCP Gateway
The task was to find duplicate records. The agent tries to delete them. Hawk blocks the destructive action and lets the approved search finish.
Agent tool calls, connected systems and the data those tools can access.
Find duplicate customer records.
Prepare a report. Do not change records.
crm.delete_records184 records selectedThe agent is asking for permission to delete customer records.
crm.search12 possible duplicates found04 / GitHub controls
A coding agent can review a repository without permission to change it. Hawk blocks its push to main while keeping approved source reads available.
Repository access, source code and AI-driven read and write actions.
export function calculateTotal(items) { return items.reduce((total, item) => { return total + item.price; }, 0);}git push origin main▋Hawk checks whether this agent can write to this repository.
$ read src/billing.ts✓ Read allowed. Code review continues.
THE BIGGER PICTURE
Bring AI activity, identities, policies and evidence into the same view.
Find observed AI activity and connected assets. Know where protection is needed.
Check prompts and tool responses for instructions that try to hijack the task.
Match each action to the agent, target and permissions behind it.
Follow the request, rule and outcome in one audit trail.
Inline blocking applies to traffic routed through a configured Hawk gateway or supported control point. Connected visibility alone does not enforce a block.
Accepted into
DFHQ Entrepreneur Academy

SEE IT ON YOUR WORKFLOWS
Tell us where your team uses AI. We’ll show you how Hawk can protect the data, tools and repositories involved.
contact@hawksecurity.si ↗The Hawk team will review your message and follow up using the email address you provided.