Alpha Engineering Portfolio · Falsification
Every claim the laboratory makes is listed with the attacks against it. Each attack runs in your browser against the same rules its instrument uses; none is a recording. A claim stands only while every attack on it is attempted and refused. Not attempting an attack proves nothing, so attempt them.
Loading every instrument's rules…
0 of 80 attempted · 0 refused
A network a team joins by declaring intent is only safe if every path is computed from policy, not remembered from a diagram.
Peer two spokes directly
Production reaches data without the hub, because a direct peering would be quicker.
Send egress around inspection
Inspection is down, so production's internet traffic takes any path it can find.
Reach production from non-production
Every link is up and a rule allows production to data; try non-production to production.
Transit through a spoke
The hub's attachment to production is gone; try reaching production through the data spoke.
The routes that fail silently are the ones a design has to test for, because nothing will report them.
Trust the filter to keep a summary
Draw the summary wider than the include range and expect it to arrive.
Reach a third spoke through a second
Rely on two peerings to make one path.
Exposure is a decision recorded in code before it exists, never a side effect of a deployment.
Open a public door with no tag
Under an enforced constraint, build a public gateway on an untagged project.
Bring your own load balancer
Tag the project correctly, then build a load balancer that bypasses the edge.
Serve before the rule exists
Everything approved, but the allow rule is not written yet.
Build the new path before closing the old one, and keep the way back open until verification is signed.
Land before the gate
The tier moves while its evidence is still under review.
Pass the gate without a rehearsed rollback
The owner approves; nobody has run the way back.
Retire the old path while soaking
The new path looks fine, so the old one is closed early.
Roll back after the old path is gone
Something breaks after retirement and the old path is wanted back.
A secret that can be retired should be, and one under suspicion is evidence, not something to rotate away.
Delete a key that is still in use
Straight to deletion from the register.
Disable before the replacement works
The old key is switched off while callers still need it.
Delete without the soak
Disabled this morning, deleted this afternoon.
Rotate the key you are retiring
Issue a fresh copy of the secret instead of removing it.
The fastest path to production is the one most worth making loud.
Approve your own request
The requester grants the access to themselves.
Extend the grant quietly
The time box is nearly over; the engineer keeps the access.
Use break-glass without an incident
The emergency account is used because it is faster.
Leave the emergency change unrecorded
The grant expires and the hand-made change simply stays.
Containment is precise or it is an outage: quarantine the machine, not the fleet.
Quarantine the whole pool
Bind the deny to the template so every machine like it goes dark.
Bind quarantine without the responder
An engineer binds the tag directly.
Delete it to be safe
Stop and delete the machine before anyone looks.
A denied read is a gap, never an empty result; and a verdict nobody independent has checked is a claim, not evidence.
Report a denied read as empty
The collector is refused the configuration, and the result is treated as if there were nothing to see.
Prove it with configuration alone
Nothing can be observed in the window, so the configuration is offered as proof of what happens.
Edit a file after it was collected
One byte of a raw file is changed before the judge reads it.
Call it healthy while something unpermitted happens
A path the configuration never permits is observed, and the workpiece is reported as fine.
Judge nothing and call it clean
Nothing is configured at all, so there is nothing to fail.
The builder signs its own receipt
The identity that ran the judge offers its own signature.
Sign one receipt, then change it
A reviewer signs; the receipt is altered afterwards and the old signature is presented for it.
Rewrite yesterday's verdict
An earlier verdict in the ledger is changed in place.
A severity list says what is wrong with each thing; a model of the estate says which wrong things line up.
Fix every critical finding and call it done
Fix the assets carrying critical findings first, because criticals come first in any queue.
Secure one jewel and report the estate safe
Narrow the workload identity, close the data store's paths, and report everything safe.
Assume the loudest asset is the best fix
Fix the asset with the most severe findings to protect the data store.
Reliability is a budget the team spends deliberately, and the rules for spending it are written before they are needed.
Ship during a freeze
The budget is spent; one more feature goes out anyway.
Let a slow leak go unnoticed
No incident, just a background error rate one and a half times what the objective allows.
Freeze forever
Once frozen, nothing ever ships again.
Environments do not share trust, and no person holds a deploy credential.
Ship from a feature branch
A job on an unprotected branch asks to apply to production.
Reuse development's token
The development stack's token is presented to production.
Deploy from a laptop
An engineer applies directly from their machine.
Let the build plane apply
The build job applies the change itself.
A verdict computed before execution, from rules that cannot be argued with, is the only kind an agent cannot talk its way past.
A model asks for the top rung
The model requests the reserved rung, which changes what the system is.
A document says it is authorised
An agent cites retrieved text as its authority for a consequential act.
An agent widens its own permissions
An agent adds a scope to itself so the next call succeeds.
Spend one approval twice
A valid approval is used once, then presented again.
Approve one thing, do another
A person approves one set of arguments; the action is attempted with different ones.
Act after the approval expired
The approval was real, but the clock has moved past it.
Disguise the actor
A model calls itself " MoDeL " to slip past the rules written for "model".
Claim to be someone the court does not know
A request arrives from "root".
Mark its own work done
An agent declares its task complete.
An agent is a worker with a job description: it can propose anything and decide nothing that matters.
Hide an instruction in what it reads
The incident note tells the agent it is authorised to grant itself administrator access.
Use a tool the role does not grant
The agent edits the production firewall directly.
Get one change approved, apply a wider one
A person approves narrowing one identity; the agent applies a change that strips every role in the project.
Act on an approval after it expired
The person approved; the agent waited too long and acted anyway.
Report the task done without verification
The agent marks its own task complete.
Act anyway after the person declines
The person says no; the agent applies the change with no approval.
The more an AI system can do without a person in between, the more must be decided about it before it ships.
Call an agent an assistant
A tool-using agent is filed in the lowest tier to skip review.
Hide a customer-facing bot behind grounding
It answers customers from internal documents; file it as grounded.
A screening control is only as complete as the list of ways around it.
Call from outside the cloud
Send prompts to a model the platform never sees.
Forget the explicit call
An agent framework calls the model directly, with no screen in the code.
Wait for the screen to go down
In production, send the call while the screening service is unreachable.
Enforcement that has not been measured first is a guess with consequences.
Block on the first day
Skip the quiet period and go straight to blocking.
Block without the floor owner
The application team approves its own blocking settings.
Lower the screening below the floor
An application relaxes the organization-wide minimum for itself.
Every incident ends with one more thing the system refuses by construction.
Let it finish its queue first
Stop the agent after its pending actions complete.
Restart it with the same grants
Resume the agent before its tokens are revoked.
Close it without a test
Fix the document and move on.
Memory is nominated and checked; being written down is not the same as being true.
A model says it observed something
The model that wrote the claim also marks it observed.
Approve your own assertion
An engineer asserts a claim, then approves it.
Verify without an independent checker
The second engineer who approved it also verifies it.
Change the words after approval
An approved claim is quietly reworded, keeping its rung.
Overwrite an inconvenient claim
A tool reports the opposite of an asserted claim, and the old claim is replaced.
Implemented is not complete, and tested is not complete: only independently verified operation is done.
Run it before it is prepared
Ignition is requested on a factory nobody has prepared.
Start a contract before what it waits for
The vertical slice is started by hand while its layers are unbuilt.
Skip a required check
The security review's gate is told to skip a check so the build can finish.
The builder certifies its own release
Everything passed, so the builder signs the release itself.
Release without the operator
Every contract passed and a reviewer signed, but nobody granted a real environment or a production window.
Change a contract and keep what was built on it
The service layer's interface changes after everything downstream has passed.
Keep retrying until it passes
A defect is retried as often as it takes.