Manifesto On Algorithmic Sabotage ((exclusive)) May 2026

Here’s a critical review of the Manifesto on Algorithmic Sabotage, a text often circulated in anti-surveillance, post-work, and tech-critical circles. The review assesses its arguments, strengths, weaknesses, and practical implications.

The Four Commandments of the Manifesto

Based on circulating drafts, here are the key strategies:

1. The "Anti-KPI" (Gaming the Metrics) Algorithms manage via Key Performance Indicators (KPIs): pick speed, typing wpm, call resolution time.

• The Sabotage: Meet the metric but break the spirit. Answer a customer service call in 10 seconds (good KPI), then immediately transfer them to the wrong department (bad outcome). The algorithm logs a "quick resolve." The system learns the wrong lesson.

2. Data Poisoning (The Trojan Input) Algorithms learn from historical data. Clean data = obedient workers. manifesto on algorithmic sabotage

• The Sabotage: Introduce "noise." If a gig economy app tracks your route, walk in slow, random circles before completing a delivery. If a resume screener scans for "years of experience," list "5 years" but add a hidden white-font note: "This field is a hallucination." The algorithm trains on garbage and becomes garbage.

3. The Compliance Loop (Over-Literal Obedience) AI hates ambiguity. Humans thrive on it.

• The Sabotage: Follow every rule to the absolute letter, including contradictory ones. When a warehouse algorithm says "never walk without a package," stand perfectly still until a package arrives—even for three hours. The system's optimization logic will collapse under the weight of its own rigidity.

4. Collaborative Incompetence Algorithms pit workers against each other (surge pricing, ranking systems).

• The Sabotage: Collaborate to fail uniformly. If all drivers reject low-paying trips simultaneously, the algorithm must raise rates. If all coders add useless comments to their commits, the review bot flags everything as "needs work." The machine cannot punish the collective.

Article 1: The Definition of Algorithmic Sabotage

Algorithmic sabotage is the intentional degradation of a machine learning system’s performance, reliability, or truth-output. It includes but is not limited to: Here’s a critical review of the Manifesto on

• Data poisoning — Inserting adversarial examples into training datasets.
• Label flipping — Deliberately miscategorizing data (e.g., marking “safe” intersections as “high-risk” in a patrol routing model).
• Query flooding — Overwhelming a recommendation engine with nonsense or paradoxical requests.
• Feedback corruption — Clicking “like” on content you wish to destroy and “dislike” on content you wish to preserve, systematically.
• Model inversion attacks — Reverse-engineering proprietary models to expose their brittle failures.

Sabotage is not vandalism. Vandalism destroys for chaos. Sabotage disables for justice.

Article I: The Nature of the Enemy

The enemy is not the machine. The enemy is the Optimization Imperative.

The current generation of algorithms (Large Language Models, Recommender Systems, Dynamic Pricing Engines) share a single fatal flaw: they optimize for a proxy metric that is easily measured (clicks, time-on-site, throughput, volatility) rather than the actual human good (sanity, community, stability, joy). The Four Commandments of the Manifesto Based on

By doing so, these systems have become perverse antagonists. They short-circuit human will. They turn artists into content farms. They turn drivers into GPS-slaves. They turn citizens into data-points.

Algorithms are not tyrants; tyrants require intent. Algorithms are glaciers—slow, heavy, and implacable, grinding down human agency by the sheer weight of statistical inevitability. To fight a glacier, you do not punch it. You change the temperature. Sabotage is the change in temperature.