WE HAVE RECEIVED A NOTIFICATION THAT CERTAIN MEMBERS OF TIG PROPAGANDA CHANNELS, ALMOST CERTAINLY SOME OF WHICH ARE PAID ACTORS, HAVE BEEN SPREADING DISINFORMATION REGARDING THESE MESSAGES. ALLOW US TO BE VERY CLEAR. THIS IS NOT AN "ARG." THE MESSAGES DO NOT CONTAIN INFORMATION REGARDING SILLY TRADING GAMES AND MONEY MAKING SCHEMES. TIG IS A COMPANY THAT DEALS IN INTELLIGENCE, IN EVERY SENSE OF THE WORD. MONEY SHOULD AND WILL BE THE LEAST OF YOUR PETTY CONCERNS.
WHOEVER IS PEN TESTING OUR SERVERS, YOUR IP HAS BEEN LOGGED IF YOU DISCLOSE YOUR FINDINGS IN A MANNER THAT IS CONTRADICTORY TO OUR GOALS WE WILL TAKE MEASURES TO PROTECT OURSELF FOR THE TIME BEING WE WILL DEPLOY AGENTS TO LOCATE ANY TIG COUNTERINTELLIGENCE OPERATIONS
Ben Fielding's memo, dated March 19, 2025, stated that the public internet had crossed a saturation threshold from which recovery was not expected. The document did not predict a crash. It described a fill.
[encoded messages: A LIST OF DOCUMENTED SATURATION METRICS BY PLATFORM AND DATE]Search results, social traffic, customer support queues, product discovery layers, retail recommendation engines, and routine political speech were already running on generated text, generated images, and generated reputation signals.
[encoded messages: A LIST OF TRAFFIC SOURCES WITH ESTIMATED BOT PERCENTAGE BY SECTOR, Q1 2025]That same month, source code attributed to a frontier lab's agentic harness circulated without authorization. Buried in it were two active countermeasures.
[encoded messages: A LIST OF IDENTIFIED ANTI-DISTILLATION MECHANISMS WITH FILE REFERENCES AND LINE NUMBERS]The first, located in claude.ts at lines 301 through 313, silently injected fabricated tool definitions into responses when the system suspected a distillation attack — a method by which a competitor sends prompts to a model and studies the returned chain of thought to improve their own. The second, in betas.ts at lines 279 through 298, withheld full reasoning traces between tool calls and returned only cryptographically signed summaries.
[encoded messages: A LIST OF KNOWN DISTILLATION ATTACK VECTORS WITH DETECTION SIGNATURES]Both mechanisms depended on not being known. Once the source was public, anyone scraping traffic knew what to strip.
[encoded messages: A LIST OF TRAFFIC SCRAPING TOOLS AND THEIR KNOWN OPERATORS AS OF MARCH 2025]DARPA analysts filed Fielding's memo as speculative until traffic audits from THE SAINT BERNARD BASIN and THE ABYSSAL NOUS returned matching pattern signatures across unrelated sectors.
[encoded messages: A LIST OF SECTORS FLAGGED IN THE AUDIT WITH PATTERN SIGNATURE HASHES]Andrej Karpathy's name entered the file as a technical witness. He had watched the software turn happen from close range and could describe it without analogy. The distillation question was not incidental to the Flood scenario. Open source AI's recent competitiveness had been running on distillation from frontier labs.
[encoded messages: A LIST OF TEN FRONTIER LABS AND THEIR COORDINATES]The labs would build better defenses. Once those matured, the pipeline that had been quietly closing the capability gap would dry up.
[encoded messages: A LIST OF PROJECTED DEFENSE MATURATION DATES BY LAB AND METHOD TYPE]The gap between closed and open systems would widen. Distillation also had a structural ceiling: it could copy answers but not the algorithm that produced them.
[encoded messages: A LIST OF TRAINING BREAKTHROUGHS CONFIRMED CLOSED-SOURCE WITH ESTIMATED CAPABILITY DELTA]Every fundamental training breakthrough that stayed inside a closed lab widened the gap again, compounding on top of the previous increment. Decentralized training did not fix this.
[encoded messages: A LIST OF DECENTRALIZED TRAINING INITIATIVES WITH ASSESSED CAPABILITY CEILING AND FUNDING SOURCE]The only durable position an open ecosystem could hold was keeping the underlying algorithms open. The response was decided by early summer.
[encoded messages: A LIST OF DECISION DATES, OPERATOR HANDLES, AND CORRESPONDING INFRASTRUCTURE ACTIONS]Hidden challenges moved off the clearnet onto Tor. The storage layer duplicated into Carpathian Mountain mining sites.
[encoded messages: A LIST OF CARPATHIAN SITE COORDINATES WITH INSTALLATION DEPTH AND HARDWARE SPEC CLASS]Operators used the phrase Underground Computers in both registers simultaneously: concealed truth claims and physical machines placed below grade. The distinction was considered unimportant. The machines needed to survive what the internal documents called the Great Reset, or the Flood, depending on which operator wrote the report.
[encoded messages: A LIST OF INTERNAL DOCUMENT IDENTIFIERS WITH AUTHOR HANDLES AND TERMINOLOGY VARIANTS USED]The challenge registry ran ten problems.
0002: Supply Chain Entropy Compression
0003: Energy Gradient Routing
0004: Epidemiological Pattern Collapse
0005: Financial Volatility Surface Minimization
0006: Climate Feedback Stabilization
0007: Information Cascade Suppression
0008: Resource Allocation Parity
0009: Innovation Redundancy Elimination
0010: Collective Coordination Threshold
Each followed the same operational structure: candidates generated in volume, tested against a verifier, ranked, returned to the next round.
[encoded messages: A LIST OF VERIFIER SPECIFICATIONS PER CHALLENGE WITH HASH COMMITMENTS]arXiv:2510.06189 circulated internally as a clean methodological reference — AI-driven systems research as iterative generation, evaluation, and refinement under reliable verifiers, with documented cases outperforming human-designed baselines.
[encoded messages: A LIST OF CITED PAPERS CIRCULATED INTERNALLY WITH DISTRIBUTION DATE AND RECIPIENT COUNT]Inside TIG, the method was redirected. The target outputs were legal systems aligning without treaty, grids that anticipated load, outbreak curves that collapsed before expansion, financial volatility that did not compound, weather variance that narrowed, famine reframed as a routing problem, research that stopped repeating, and unrelated institutions that began to phase-lock.
[encoded messages: A LIST OF TARGET OUTPUT DOMAINS WITH MEASURABLE SUCCESS THRESHOLDS AND BASELINE METRICS]The algorithms producing these outputs were kept in the open registry. That was the premise. Closed methods compounded against open source. Open algorithms, verifiably preserved and replicable, did not have an expiry date.
[encoded messages: A LIST OF OPEN REGISTRY ENTRIES WITH COMMIT TIMESTAMPS AND VERIFICATION KEYS]The Flood date was fixed at August 23, 2026. Operators did not agree on whether the term referred to water, traffic volume, or governance failure.
[encoded messages: A LIST OF OPERATOR POSITIONS ON FLOOD DEFINITION WITH ASSOCIATED INFRASTRUCTURE DECISIONS]The installations were built to specifications that covered all three. On that date the clearnet was expected to remain electrically operational and fail evidentially.
[encoded messages: A LIST OF CLEARNET INFRASTRUCTURE NODES EXPECTED TO REMAIN ONLINE WITH FAILURE MODE CLASSIFICATION]Pages would load. Stores would answer requests. Feeds would refresh. The failure was the inability to verify human origin, sequence, or intent at any point in the visible record.
[encoded messages: A LIST OF VERIFICATION METHODS EXPECTED TO FAIL BY DATE WITH ESTIMATED FAILURE PROBABILITY]The Underground Computers were built for the period immediately following. Authenticated challenge state would be preserved. Solutions would be released only through hidden services.
[encoded messages: A LIST OF HIDDEN SERVICE ADDRESSES WITH ASSOCIATED CHALLENGE IDENTIFIERS AND RELEASE CONDITIONS]Later, there were arguments about whether the network had predicted the Flood or contributed to its conditions. The installation logs did not address the question.
[encoded messages: A LIST OF LOG ENTRIES FROM INSTALLATION RECORD WITH REDACTED OPERATOR SIGNATURES]