AN ORDINARY TUESDAY · UNMANAGED FLEET
It's already happening on your laptops.
CLAUDE DESKTOP · ENG
CURSOR · PLATFORM
CHATGPT · SUPPORT
COPILOT · ENG
CLAUDE CODE · INFRA
CLAUDE COWORK · SUCCESS
VS CODE · DATA
NONE OF IT WAS SEEN
47 days
average time exfiltration goes undetected (Verizon DBIR)
89%
of enterprises don't know who uses which AI tools (Gartner, 2026)
60%
of AI IDEs route straight past the system proxy; native binaries ignore HTTPS_PROXY
$4.9M
average cost of a breach (IBM, 2025) · a single leak dwarfs the cost of prevention
WHY KERNEL-LEVEL
Your proxy never saw any of it leave.
Native binaries ignore HTTPS_PROXY. CLIs open direct TLS connections. Web apps route around appliances inside the browser sandbox. If enforcement lives above the operating system, AI traffic simply goes underneath it.
The stack behind the terminal.
THREE LAYERS · NOTHING ROUTES AROUND
01
Kernel interception
A Network Extension (NE) sees every TCP connection at the OS layer and whitelists only AI-app hosts. Apps can't route around it.
02
Transparent in-line inspection
AI-app TLS is decrypted on-device and parsed — JSON, gRPC, protobuf — while everything else tunnels through untouched.
03
Content verdicts
Credit cards, API keys, secrets and your custom patterns are blocked before transmission — and every verdict is logged.
PRIVACY & TRUST
ON-DEVICE · NOTHING STORED
Inspection without surveillance.
Only whitelisted AI-app hosts are decrypted, on the device itself. Employees are informed; enforcement is policy-based. Audit logs carry verdicts, never content.
All other traffic tunnels through fully encrypted
No browsing history, email or personal apps
No prompt payloads stored anywhere
TEAM COLLABORATION
EVERY PARTICIPANT COVERED
Shared sessions, covered.
Claude Cowork multiplies exposure: shared prompts, shared artifacts, one-click uploads, no per-user trail. Aegis enforces policy for every participant at the network layer and attributes every attempt in the audit log.
Why security teams pick Aegis.
PURPOSE-BUILT FOR AI TRAFFIC
Prevents, not just detects
Sensitive data is blocked before it reaches the LLM, not flagged after.
Unbypassable by design
Kernel-level enforcement on managed Macs; not a proxy, not an agent, and users can't disable it.
Full Claude ecosystem
Desktop, Code and Cowork, plus gRPC/protobuf parsing for Cursor.
Invisible to users
No VPN, no MDM, no install wizard. Developer ID native on macOS.
NE architecture proven at Bitdefender-class scale
gRPC/protobuf parsing on live Cursor traffic
100+ concurrent flows · zero user friction
DEPLOYMENT
ROLLOUT IN MINUTES
Runs quietly in your fleet.
Developer ID-signed installer; MDM optional for fleet-wide push
macOS today, native NE; Windows agent on the roadmap
Fail-safe by policy: AI-app traffic fails closed; everything else unaffected
Your definitions: custom patterns for internal key formats, codenames, PII
COMPLIANCE
EVIDENCE, NOT PROMISES
The trail your auditors ask for.
GDPRArt. 32 technical measures; PII blocked before it leaves the device
SOXPer-user, per-app enforcement with tamper-evident logs
HIPAAPHI patterns detected and blocked across all AI apps
SIEMEvery flow is a logged, attributable verdict; exports to Splunk or Sentinel
Two weeks to full visibility.
NO IT INFRASTRUCTURE NEEDED
DAY 1
Install on a pilot group
Signed installer on ten machines; no proxy config, no VPN profile.
WEEK 1
Shadow AI visibility report
Who uses which AI apps, and what nearly leaked; observe-only mode.
WEEK 2
Enforcement on
Blocking and audit logs across the pilot; expand to the fleet when ready.
The questions your red team will ask.
STRAIGHT ANSWERS
Can a user just uninstall it?
Not without admin approval — removal requires elevated rights and alerts your security team. Users can't disable, pause or route around the extension from their session.
What does the employee see when something is blocked?
An inline notice with the exact policy reason — no silent failures, no mystery errors. The rest of their session keeps working.
Does this slow the machine down?
Only AI-app flows are inspected; everything else tunnels straight through. Tested at 100+ concurrent flows with no visible latency.
What about personal devices and hotspots?
Aegis protects managed devices — enforcement travels with the machine, on any network, VPN or hotspot. Unmanaged personal hardware is a policy conversation, not a technical one; the pilot's visibility report shows you how much of your exposure is on managed machines today.
Are you reading our prompts?
No. Inspection happens on the device and nothing is stored — audit logs carry verdicts ("blocked: aws-access-key"), never content. See Privacy & Trust.