Stratagems #20: Alex Felt the AI Collector Slow Down. He Knew Someone Else Had Made a Move.

Don't fight in clear water — stir it first. Don't strike when they're watching — strike when they're looking somewhere else.
— The 36 Stratagems, Disturb the Water and Catch a Fish

Previously on this series:

#17: Alex Set a Bait. The Catch Wasn't Code — It Was Someone Who Shouldn't Have Been Watching. — Alex planted a shadow training pipeline inside MedTech, let ACL's extraction pipe pull watermarked data. Found another set of eyes in ACL's infrastructure — a PGP key and an encrypted channel left by an unknown operator. He sent a carefully crafted data packet, a digital breadcrumb, designed to be subtly different from legitimate training data. This packet contained watermarks, invisible to the human eye but detectable by AI, meant to trace the origin of any unauthorized use. The goal was to identify who was monitoring MedTech's AI development, specifically within the Advanced Cognitive Labs (ACL) infrastructure.

The initial investigation in Stratagems #17 yielded critical intelligence: a PGP key and an encrypted channel indicated a sophisticated operator, not a casual eavesdropper. Alex had confirmed unauthorized access, but the identity and motive remained obscured. This new development, however, suggests the operator is now actively engaged in data exfiltration, moving beyond mere surveillance.

The Subtle Signal of a Slowdown

Alex's AI data collector, a sophisticated system designed to ingest and process vast amounts of information for training machine learning models, began exhibiting an anomaly. It wasn't a critical failure or a complete halt; rather, it was a perceptible slowdown. The system, usually operating with predictable efficiency, now showed a consistent, albeit minor, degradation in processing speed. This kind of subtle performance dip, easily dismissed as background noise or routine system load, immediately raised Alex's suspicion. In the high-stakes world of AI development, where every millisecond can impact model training and competitive advantage, such deviations are not random. They are signals.

Think of it like a highly trained athlete suddenly feeling a slight fatigue during a routine sprint. The athlete doesn't stop; they might even complete the sprint. But the experienced coach, observing the subtle change in gait and breath, knows something is amiss. Alex, in this analogy, is the coach. He understood that the slowdown wasn't a system glitch but an external influence.

The slowdown was directly correlated with the data flowing through the ACL's extraction pipeline. This pipeline was Alex's honeypot, designed to lure and track unauthorized access to MedTech's proprietary data. The watermarked data Alex had seeded was being pulled, but now, something else was happening: the sheer volume or the nature of the extraction was impacting the collector's performance. This implied that the unknown operator, previously just a pair of eyes, was now actively and aggressively siphoning data, creating a bottleneck that even Alex’s sophisticated infrastructure could detect.

Identifying the Adversary's Strategy

The stratagem at play here is a variation of "Disturb the Water and Catch a Fish." The unknown operator isn't directly confronting Alex or MedTech. Instead, they are subtly disrupting the environment – the data pipeline – to achieve their objective: data theft. By creating a demand on the extraction resources, they are essentially forcing the water to churn, making it easier to scoop out the valuable data without being immediately detected by standard monitoring tools that look for overt intrusions.

Alex's previous discovery of the PGP key and encrypted channel pointed to a well-resourced and technically adept adversary. This slowdown confirms their active engagement. The operator is likely attempting to exfiltrate a significant portion of the watermarked training data, perhaps to reverse-engineer MedTech's AI models, gain insights into their research direction, or even to use the data for their own competing AI development. The fact that they are operating within ACL's infrastructure, using its resources, suggests either a compromised insider or a sophisticated external actor who has gained deep access.

The critical insight is that the operator is not merely observing; they are acting. This transition from passive surveillance to active extraction escalates the threat significantly. Alex needs to understand not just *that* his data is being taken, but *how much* and *what specific components* are being targeted. This information will be crucial in determining the extent of the damage and formulating a counter-offensive.

The Counter-Move: Adapting to the New Threat

Alex's next step must be to leverage this newfound knowledge. The slowdown is the anomaly, and anomalies are opportunities. He cannot simply shut down the pipeline, as that would alert the adversary and potentially lose the trail. Instead, he must adapt his strategy.

One immediate action is to enhance the watermarking and tracking mechanisms. If the operator is pulling large volumes, Alex can embed more granular watermarks, perhaps even dynamic ones that change based on the query or extraction pattern. This would allow him to identify not just the source of the theft but also the specific datasets being targeted, providing a clearer picture of the adversary's interests.

Furthermore, Alex can use the detected slowdown as a lever. By understanding the extraction rate, he can calculate the approximate time it would take for the adversary to acquire a significant amount of data. This allows him to prepare a response, whether it's deploying countermeasures, alerting relevant parties, or initiating a disinformation campaign. The principle of "strike when they're looking somewhere else" can be inverted; Alex now knows where the adversary is looking, and can plan his move for when they are maximally engaged in their current operation.

The situation has evolved from a reconnaissance mission to an active espionage operation. Alex's ability to detect the subtle slowdown is a testament to his meticulous monitoring and understanding of his systems. Now, the challenge is to turn this detection into decisive action, using the adversary's own aggressive moves against them.