Airlock Digital · Airlock 7.1 launch · pre-messaging brainstorm

Agentic Steering · Concept Matrix

Primary campaign message

Version 7.1 extends Airlock Digital Application Control to actions initiated by supported agentic agents. Security teams can see the commands agents attempt, evaluate those actions against policy before execution, and steer supported agents with the reason for a denial or an approved alternative where configured.

Supporting message

Application Control governs what software can run, while Agentic Steering extends that same preventative policy foundation to what supported agents can execute after they are running. Version 7.1 helps security teams adopt agentic AI with greater confidence by providing visibility into agent activity, policy-based control before execution, and actionable guidance when an action is denied or redirected.

Stay true to the product
  • Each agent command checked before it runs
  • Allowed or denied by the customer’s policy
  • Reason or approved alternative returned to supported agents, where configured
  • Sessions, commands, decisions and outcomes logged
Starred from this page

Short list

Nothing starred yet. Shared: everyone who opens this page sees the same list.

01

Territory matrix

Twelve worlds, same eight rows. Nouns and verbs are for word-hunting; headline starters are finished lines; the watch-out is the trap in each world.

02

Verb × object grid

A line generator. Read a verb across its objects; blank cells are weak pairings, left empty on purpose. Scrolls sideways on small screens.

verbthe agentthe commandthe actionthe routethe risk
steerSteer the agent. Keep the speed.Steer the command before it runs.Steer the action, not the aftermath.Steer around the risk.
clearClear the agent to work.Cleared, command by command.Clear the action. Log the outcome.
redirectRedirect the agent, with the reason.Redirect to a command you’ve approved.Redirect the risky move.
seeSee what the agent attempts.See every command, even the ones that never ran.See the action before it happens.See risk while it’s still a request.
correctCorrect the agent mid-task.Correct the command. Keep the task.Course, corrected.
approveApprove commands, not whole agents.Approve the action. Keep the receipt.Approved routes only.
rerouteReroute the agent, not the project.Reroute to the approved command.Reroute, don’t roadblock.
guideGuide the agent with a reason, not a wall.Guide every action by policy.Guide the route, turn by turn.
checkCheck the agent at every step.Check the command. Then let it run.Checked before action.Check the risk before the run.
rewriteHand back the approved rewrite.Rewrite the route. Keep the destination.
denyDeny the shortcut, not the task.Deny the command. Explain the denial.Deny the action. Offer the alternative.Close one route. Point to another.Deny risk by default.
logLog every agent session.Log every command, run or not.Every action, on the record.
03

Mash-ups

Two territories in one line. The pair is labelled so either visual can carry it.

04

Brightbase picks

Twelve lines that survive the red pen, grouped by the job they do best.

Strongest for the steering-wheel visual

Lines that give the wheel a second party: the policy.

01

Your agents drive. Your policy steers.

Splits the roles in six words and lands on the product’s own verb.

Steering and driving
02

The wheel is theirs. The doors are yours.

Puts the brand name inside the steering idea without a pun.

Steering × Airlock
03

Lane assist for AI agents.

Everyone has felt lane assist: a nudge back, not a stop. That is the redirect.

Steering and driving
04

Every agent needs a co-driver.

Adds the missing figure to the wheel image and hints at rules written in advance.

Wildcard: the co-driver

Strongest for the maze visual

Lines that let the dots be clever and the path be yours.

05

The agent hits a wall. The wall gives directions.

Dramatises the returned reason and alternative, the part no rival line can claim.

Mazes and paths
06

Agents try every door. Policy decides which ones open.

Honest about how persistent agents are, without fear, and nods to the airlock’s doors.

Mazes and paths
07

Agents explore. Policy draws the walls.

Gives the agent credit and the customer authorship; sits cleanly over dots in a maze.

Mazes and paths
08

Trial and error, without the error.

A knowing line for engineers; body copy must land “checked before it runs”.

Mazes and paths

Strongest typographic lines

Lines that work alone, set big, no image needed.

09

Agents ask permission, not forgiveness.

Flips a known idiom into the exact behaviour: checked before it runs.

Permission and trust
10

Checked before it runs. Not cleaned up after.

Speaks to incident fatigue with a plain claim the product can prove.

Before it runs
11

Guardrails stop. Guide rails steer.

The steering-beats-braking argument in five words; strong with analysts.

Coaching and guidance
05

Taglines and sign-offs

End-lines that sit under any execution.

  1. Agentic Steering, new in Airlock 7.1.
  2. Allow. Deny. Steer. Airlock 7.1.
  3. See it. Control it. Correct it.
  4. Deny by default. Steer by policy.
  5. Your policy speaks first.
  6. Your policy, at every command.
  7. Agents welcome. Policy first.
  8. Checked before it runs.
  9. Airlock 7.1. Now with directions.
  10. Application control, now for AI agents.

Brainstorm draft, not approved copy. Before any line ships, confirm with Airlock which agents are supported and where approved alternatives are configured; lines that imply a reason or detour on every denial need that qualifier in body copy.