Running Silent

Bottom Line

The direction is right. The physics is wrong.

A widely circulated AI-generated text claims that ambient radio, commercial transponder feeds and machine learning can be fused into a real-time global map of hidden submarines and carrier groups, with no decryption required. This briefing tested each assertion against open physics, peer-reviewed research and programme records.

What actually works is narrower, more expensive and more active than the text implies: satellite radar matched against AIS, satellite geolocation of ships that still emit, seabed and towed acoustic arrays, skywave over-the-horizon radar cued to imaging satellites, and fusion software now used by more than a hundred thousand US personnel. None of it makes running silent obsolete. It makes running silent insufficient on its own.

How every claim was graded

Key Findings

Eight judgments, with confidence stated

1. 01
   The VLF submarine mechanism is wrong. Seawater conducts. VLF fields fall by about 5 dB per metre of depth and a hull is a thousandth of a wavelength. Any scattered field sits hundreds of decibels below lightning noise. Submarines receive VLF; shore stations transmit it.
   HIGH
2. 02
   Real non-acoustic submarine detection is short-range or experimental. Magnetic anomaly detection works at hundreds of metres to low kilometres. Chinese magnetometer advances reported through the South China Morning Post are genuine laboratory results, not wide-area search capabilities.
   MOD–HIGH
3. 03
   Passive acoustics is the one mature “passive” layer. The US undersea surveillance system, SURTASS and TRAPS are real and funded. China’s “Underwater Great Wall” is partly deployed. Telecom cables now detect ships through distributed acoustic sensing in peer-reviewed trials.
   HIGH
4. 04
   Dark-vessel analytics are mature, but the text’s logic is inverted. Detection needs an independent sensor first. Warships are generally not required to broadcast AIS, and Russian AIS and GNSS spoofing is now routine enough that fourteen European states issued a joint warning in January 2026.
   HIGH
5. 05
   Passive radar is real and range-limited. Broadcast-illuminated systems track aircraft to a few hundred kilometres near transmitter-rich land. Starlink- and GNSS-illuminated passive radar reaches single-digit kilometres in the literature.
   HIGH
6. 06
   The real over-the-horizon threat to carriers is active and dedicated. Skywave radar cues coarsely at long range; imaging and electronic-intelligence satellites refine the fix. Western analysts still describe the full Chinese anti-ship missile chain as untested against a defended, manoeuvring carrier.
   MOD–HIGH
7. 07
   Fusion software is scaling fast; a real-time global “glass map” is not here. The Maven Smart System ceiling is about $1.3 billion through 2029. It fuses existing intelligence feeds. Palantir’s “Gaia” is a map interface, not a detection engine.
   HIGH
8. 08
   “Running silent is no longer a defense” is overstated. Emission control plus dispersion and deception still impose large search costs. Both the US and its competitors keep investing in counter-surveillance, which is what you do when concealment still pays.
   MODERATE

Scope and Method

Two very different things share one phrase

“Over-the-horizon target fusion” names a genuine military capability and a largely fictional one. This briefing keeps them apart.

Sources and their weight

Peer-reviewed literature (Nature, JASA, IEEE, IET, Sensors, arXiv), US government records (NRO releases, DARPA programme pages, Navy budget exhibits, Naval History and Heritage Command), Palantir’s own technical documentation, and established defence press (C4ISRNET, Defense News, Breaking Defense, DefenseScoop, The War Zone, SpaceNews). Chinese claims are attributed to their reporting outlet, usually the South China Morning Post summarising Chinese-language journals, and are treated as claims.

The original text opens by congratulating its reader for having “independently architected” a classified programme. That flattery is set aside. The technical substance is judged on its merits, sentence by sentence, in the verification ledger at the end of this document.

Chapter 1 · VLF physics and real non-acoustic detection

Why seawater ends the VLF story in two metres

The claim: passive sensors watch VLF waves “warp or deflect around a massive metallic object under the water” and locate a submarine that never transmits. Grade (d), wrong, high confidence.

Who transmits. VLF, 3 to 30 kHz, is a one-way broadcast from very large shore stations to submarines, which receive it near the surface through towed wire or buoy antennas. The premise that submarines “use VLF to communicate” is true only in the receive direction.

Wavelength against hull. At the 10 to 30 kHz used for submarine broadcasts, wavelengths run 10 to 30 km. A Virginia-class hull is about 115 m long; an Ohio-class about 170 m. The hull is one hundredth to one two-hundred-and-fiftieth of a wavelength, deep in the Rayleigh regime, where the scattered field scales as the cube of the size ratio: about 95 dB below the incident field at the hull itself, even with the hull floating in air.

Skin depth. Seawater conducts at about 4 S/m. The skin depth at 20 kHz is about 1.8 m, so the field loses about 4.9 dB per metre. A submarine at 100 m sits roughly 480 dB below the surface field, and any scattered energy suffers the same loss again coming back up. To VLF, the sea surface is a mirror.

Noise floor. VLF ambient noise is dominated by global lightning. The hypothesised signal sits hundreds of decibels below it, and no machine-learning integration recovers a signal that is absent at the sensor. VLF methods do locate conductive bodies, but on land, to tens of metres, with the sensor almost on top of the target.

What non-acoustic detection actually looks like

Chinese claims, read against their own papers: headlines (c), papers (b)

Chapter 2 · Passive acoustic networks

The layer that is real, and the one the text left out

Seabed and towed hydrophone arrays are the mature passive sensor. Distributed acoustic sensing on telecom cables is the genuinely new one, and it does not appear in the original text at all.

United States and allies. The Cold War SOSUS seabed hydrophone system, now the Integrated Undersea Surveillance System, is the canonical “passive sea-floor chokepoint” network. SURTASS ships tow long passive and low-frequency active arrays. TRAPS, the Transformational Reliable Acoustic Path System, is a fixed passive node that exploits the deep-water reliable acoustic path from the seafloor. It began in DARPA’s Distributed Agile Submarine Hunting programme, with SAIC prototypes around 2013 and Leidos development from 2014. [Source: UPI, 28 Aug 2013; Military & Aerospace Electronics]

DARPA Ocean of Things. DARPA describes “thousands of low-cost, environmentally friendly, intelligent floats,” more than 3,000 deployed in Phase 2, with analytics that “automatically detect, track, and identify nearby vessels.” The 2018 solicitation set a ceiling of $500 per float in lots of 50,000, and the programme manager spoke of 50,000 floats covering a million square kilometres. Treat the latter as a planning figure, not a deployed one. Grade (b). [Source: DARPA programme page; BAA HR0011-18-S-0013]

China. China State Shipbuilding Corporation publicly proposed an “Underwater Great Wall” of submarine detectors in 2015–16. In January 2018 the Chinese Academy of Sciences disclosed two deep-sea acoustic sensors between Guam and the South China Sea, operating since 2016. Existence (a); effective coverage and integration with fleet operations (c), not knowable from open sources. [Source: SCMP, as reported by The War Zone, Jan 2018; National Maritime Foundation]

Cables that listen

A 2026 study in the Journal of the Acoustical Society of America used a trenched telecom cable in Norway’s Trondheimsfjord to detect and localise ships automatically, including “an unidentified ship without active AIS recording.” Chinese researchers connected distributed acoustic sensing to a 69 km optoelectronic cable off Hainan for AI-enabled vessel detection, citing “potential for detecting dark ships.” Real-time sensing has been demonstrated on 148 to 171 km of fibre without inline amplification. [Source: JASA 159(5), 2026; ResearchGate 396960937; arXiv 2509.11614]

Grade (b), moving toward (a) in coastal and littoral zones. Range from the cable is kilometres, and coverage follows cable routes, not the open ocean. The text’s “covert spy vessel dragging a sonar array” is broadly right about acoustics: surface survey and intelligence ships are mechanically noisy, run navigation radar and are large radar targets. Its “localised power signature” has no defined technical meaning.

Chapter 3 · AIS, ADS-B and dark-vessel analytics

You cannot subtract a ship that was never there

AIS and ADS-B are cooperative systems. They contain only what transmitters choose to broadcast. Removing declared traffic from a transponder feed leaves an empty set, not a map of ghosts. Grade as written: (d). Corrected version: (a).

Radar matched to AIS. Paolo and colleagues processed about two petabytes of Sentinel-1 imagery from 2017 to 2021 with deep learning and found that “72–76% of the world’s industrial fishing vessels are not publicly tracked,” along with 21–30% of transport and energy vessels. This is the gold-standard public demonstration. It is coastal-weighted, and revisit latency is days, not seconds. Grade (a). [Source: Nature 625, 2024; PMC10764273]

Radio geolocation from orbit. HawkEye 360 geolocates “marine radar, VHF radio, satellite communications, and emergency beacons” and flags geolocations with no matching AIS as “DarkRF.” Published cases include the tanker Romina going dark after Suez in 2020 and a cargo ship’s week-long AIS gap off Syria in 2021, resolved through VHF to a call at Tartus. Unseenlabs offers comparable service. Grade (a). The limit is structural: a ship with radar and radios off defeats radio geolocation entirely. Only imagery catches it. [Source: HawkEye 360, Apr 2022; Via Satellite, Jan 2021]

The feed is already poisoned

Any fusion architecture that trusts AIS inherits whatever deception is in it, and in 2026 that is a lot.

The Royal Institute of Navigation’s report of 23 January 2026, drawing on more than 100 sector experts and 300 masters, records that “in 2025, at least two collisions and groundings were reported in mainstream media linked to GNSS interference in regions such as the Baltics, Straits of Hormuz and the Red Sea.” Windward reported that “more than 1,650 vessels experienced GPS and AIS interference across the Middle East Gulf on March 7 [2026], up 55% from a week ago.” Academic tools such as SeaSpoofFinder now infer spoofing from AIS itself. [Source: RIN, Impacts of GNSS Interference on Maritime Safety, 23 Jan 2026; Windward, Mar 2026; arXiv 2602.16257]

Aviation

Military transports, tankers and surveillance aircraft often fly with Mode S and ADS-B on in peacetime for airspace deconfliction. Unfiltered aggregators such as ADS-B Exchange, and multilateration of Mode S returns, let open-source communities follow them. Combat and stealth aircraft on operations turn transponders off. The Hensoldt “F-35 tracking” episode of 2018 involved aircraft reportedly flying with radar reflectors and ADS-B on.

Chapter 4 · Passive radar and signals of opportunity

Borrowed light reaches kilometres, not oceans

Passive coherent location correlates a broadcast transmitter’s direct signal against its echoes to get bistatic range and Doppler. It is mature for aircraft near transmitter-rich land, and small-scale for everything else.

The “micro-variance” mechanism does not exist in the literature

A carrier does reflect satellite signals; that is the basis of GNSS- and Starlink-illuminated passive radar. Exploiting it requires a coherent bistatic receiver near the target, with ranges of a few kilometres, because satellite signals at the surface are weak and high signal-to-noise comes only with narrow beams or long integration. No published study shows a carrier group modulating received satellite signal strength over an ocean block in a way detectable by distributed receivers. Grade (c) bordering on (d), moderate to high confidence. Over the open Pacific there are no FM or television towers to borrow.

Chapter 5 · Skywave radar and space-based persistence

What over-the-horizon targeting really is

Skywave radar bounces high-frequency energy off the ionosphere to illuminate targets 1,000 to 3,000 km away. It is active, it is coarse, and it is the real first link in the chain the text imagines.

Australia’s JORN, the US Navy’s AN/FPS-71 ROTHR, Russia’s Konteyner and Chinese systems all work this way. Ship detection is far harder than aircraft detection: ships move slowly, and their echoes sit inside powerful sea clutter in Doppler. A US patent citing Abramovich (2014) notes that skywave ship detection “is significantly more challenging than the equivalent task for aircraft.” Resolution cells run from kilometres to tens of kilometres, and ionospheric registration error dominates position uncertainty. Skywave radar can say a large surface group is roughly there. It cannot target it alone. Grade (a). [Source: US Patent 11,953,580; Naval Applications of HF OTH Radar, ResearchGate]

Barry Watts judged in 2013 that “the Chinese have yet to conduct an end-to-end test of it against a moving target at sea,” and no open source has since documented one against a defended carrier. This chain is the real-world version of “over-the-horizon target fusion.” It is built from dedicated active and imaging sensors. [Source: Air & Space Forces Magazine, Dec 2013]

Persistence from orbit is what actually erodes concealment

National Reconnaissance Office. NROL-179 was “the fourteenth overall launch of the NRO’s multi-phenomenology proliferated architecture.” The NRO states it has “hundreds of NRO satellites on orbit,” that the architecture includes electro-optical, radar and relay satellites with support to ground moving target indication, and that it is “exceeding expectations.” Existence (a); performance (c), classified. [Source: NRO Release 5-26; NRO proliferated architecture press kit, Jan 2026; Spaceflight Now, May 2026]

Space Development Agency. The Proliferated Warfighter Space Architecture’s Tracking Layer is aimed at missiles, not ships. Golden Dome and Space Force commercial-imagery budget lines were not traced to primary documents and are not quoted. [Source: Northrop Grumman, 10 Oct 2026]

Chapter 6 · Pattern of life, anomaly detection and fusion software

Palantir Gaia is a map, not a sensor

The text names “Palantir Gaia” as the platform built to execute ambient signal mapping. Gaia exists. It is the geospatial command-and-control application inside Palantir Gotham. Grade for the text’s description: (d).

Palantir’s own documentation describes Gaia as Gotham’s map application: users load “Gaia maps,” add Foundry ontology data and create objects by drawing on the map. Its UK G-Cloud 14 service definition calls Gaia “a collaborative geospatial command and control application” that contextualises “artificial intelligence (AI) detections” from other sources. It organises fused data; it does not detect submarines. Palantir is, accurately, a central fusion vendor to the US military. [Source: Palantir Foundry and Gotham documentation; UK Digital Marketplace G-Cloud 14 service definition, Nov 2024]

James Mazol, deputy under secretary of defense for research and engineering, at DefenseTalks on 22 September 2026: “In January of this year, we had about 50,000 people using Maven. Then [Operation] Epic Fury kicks off, and now we’re up to over 100,000.” A March 2026 Feinberg memo made Maven a programme of record under the Chief Digital and AI Office; NATO bought its own instance in March 2025. Fusion at scale is (a). It ingests radar, optical, signals and video detections. It does not replace them. [Source: DefenseScoop, 22 Sep 2026; DefenseScoop, 15 Apr 2026]

Three reasons the carrier inference fails

The text’s trigger: three merchants altering course by two degrees without a weather reason, plus a satellite-signal “micro-variance,” means a carrier group under emission control. Grade (c), high confidence.

Anomaly detection is mature for cooperative traffic: AIS gaps, loitering, rendezvous, spoofing jumps. Anduril Lattice, Project Linchpin, CJADC2, Project Overmatch and DARPA’s Mosaic Warfare are real programmes for networking sensors to shooters; their budget lines were not traced to primary documents, so none are quoted.

Chapter 7 · Emission control and counter-detection

How long a carrier stays hidden, on the record

Emission control is not invisibility. A carrier is a 330 m steel ship visible to radar and optical satellites in clear weather. What EMCON, dispersion and deception degrade is identification and targeting-quality tracking, which is what a long-range missile needs.

NORPAC 1982. The Navy’s official history records that Enterprise “proved the subject of extensive Soviet air, surface, and subsurface surveillance,” including “unprecedented use of Backfire bombers.” A participant’s account says the Midway group, under strict EMCON and dispersion, flew mirror-image strikes for four days without detection while Soviet regiments raided the openly operating Enterprise. The participant account is not independently confirmed in detail. [Source: NHHC, Enterprise VIII; NavWeaps]

Earlier trials. The Haystack experiments raised simulated carrier survival in a nuclear exchange from under two hours to eight to fifteen hours, and the UPTIDE series averaged about five and a half days against cruise-missile submarines. Diesel-electric boats have repeatedly penetrated carrier screens in exercises; the often-cited 2005 Gotland encounter with Ronald Reagan was not traced to a primary source here. [Source: Naval War College Review, 2011]

“Running silent is no longer a defense” is therefore (c): overstated, moderate confidence. The defensible statement is that running silent alone is increasingly insufficient against a peer with proliferated space surveillance.

Chapter 8 · Cognitive electronic warfare and decoys

Phantom fleets are a programme, not a thought experiment

The text closes by offering to explore “cognitive EW generating realistic electronic background noise” against passive mapping. Against the text’s fictional sensors the question is moot. Against real passive sensing it is an active research line. Grade (b).

Chinese doctrine, described in Western analysis as “system destruction warfare,” treats sensor networks and kill chains as the primary target, with counter-space as a central tool. Primary PLA texts were not obtained in this cycle; this summary reflects standard Western analysis, grade (b), moderate confidence. The Russian spoofing campaign in Chapter 3 is a crude state-level example of synthetic signatures at scale.

Chapter 9 · Assessment

Each layer, sorted into what is mature and what is fiction

The limits that bind every layer

Revisit and latency: hours to days for radar imagery over most of the ocean. Coverage: acoustic networks follow chokepoints and cable routes, not the open sea. False alarms: base rates punish any anomaly detector working on the whole merchant fleet. Adversarial contamination: spoofing and decoys are already in the data. And seawater, which blocks every electromagnetic method beyond metres.

Figures this briefing stands behind

Maven Smart System ceiling about $1.3B through 2029. TRAPS follow-on $72.8M (2019) plus Congressional adds of $33.766M (FY2019) and $15M (FY2020). Comprehensive FY2025 to FY2027 lines for the other programmes were not verified and are not asserted.

Verification Ledger

Every sentence, graded

1. 01
   A capability called “Passive Multi-Domain OTH Target Fusion,” “Ambient Signal Mapping” or “Passive Global Surveillance.” Not a recognised programme or doctrinal term. OTH targeting is real and active-radar based.
   (c)
2. 02
   Physics, metadata and ML can render a real-time global map of hidden assets. Fusion is real; global real-time coverage of non-emitters is undemonstrated. Maven, NRO; Paolo et al. latency.
   (c)
3. 03
   Submarines use VLF to communicate while submerged. They receive shore-transmitted VLF near the surface; they do not transmit it. Qualified.
   (a)
4. 04
   Passive sensors map a submarine via VLF “warping” around its hull. Skin depth about 1.8 m at 20 kHz; hull far smaller than wavelength; scattered field far below lightning noise. Chapter 1.
   (d)
5. 05
   AI locates the submarine from “environmental displacement of the background wave.” No signal reaches the sensor; machine learning cannot recover absent information.
   (d)
6. 06
   A spy vessel towing an array cannot hide its engine acoustics. Surface ships are noisy; detected by arrays and cable DAS. IUSS; JASA 2026.
   (a)
7. 07
   Nor its “localised power signature.” Undefined term; no documented sensor.
   (c)
8. 08
   Sea-floor passive sonar at chokepoints tracks these deviations. SOSUS/IUSS, TRAPS, Chinese seabed sensors. Leidos/NAVWAR 2019; The War Zone 2018.
   (a)
9. 09
   The system pulls live global AIS and ADS-B. Commercial feeds are widely available. Spire, HawkEye 360, ADS-B Exchange.
   (a)
10. 10
    Filtering out declared traffic leaves “ghost signatures.” Non-broadcasters are absent from AIS; detection needs radar, RF or optical first. Paolo et al., Nature 2024.
    (d)
11. 11
    Corrected: radar and RF detections unmatched to AIS reveal dark vessels. 72–76% of industrial fishing vessels not publicly tracked. Paolo et al. 2024; HawkEye 360 DarkRF.
    (a)
12. 12
    Micro-variance in background satellite signal strength over a region reveals a hidden group. No documented mechanism; satellite-illuminated passive radar reaches kilometres. IET RSN 2023; Birmingham 2019.
    (c)/(d)
13. 13
    Merchants deviating 2° without a weather reason signal an EMCON carrier. No causal mechanism; base-rate false alarms; spoofed AIS. RIN; fourteen-state letter; Chapter 6.
    (c)
14. 14
    AI infers the carrier group and renders a high-probability estimate. Real carrier targeting uses OTH radar plus satellites and is still judged untested at sea. Air Force Magazine 2013.
    (c)
15. 15
    Palantir Gaia is built to execute this. Gaia is Gotham’s geospatial command-and-control map; it displays data, it does not sense. Palantir docs; UK G-Cloud 14.
    (d)
16. 16
    Western SIGINT clouds are built to execute this. Fusion clouds exist (Maven); no evidence of the described mechanisms. DefenseScoop 2025–26.
    (c)
17. 17
    “Running silent is no longer a defense.” Overstated; EMCON still raises search costs; both sides fund deception. Naval War College Review 2011; NHHC; NEMESIS.
    (c)
18. 18
    Cognitive EW generating realistic background noise is deployed against passive mapping. Cognitive EW and multi-sensor deception are real; framing against fictional sensors is moot. DARPA BLADE/ARC; NEMESIS.
    (b)

Conclusions and Implications

Four judgments and three things to do about them

1. KJ-1
   The text’s three most specific mechanisms are wrong or unsupported. VLF hull scattering, AIS subtraction and satellite-signal micro-variance. Its general thesis, that fusion is eroding concealment, is directionally correct.
   HIGH
2. KJ-2
   Surface-fleet concealment is being eroded by proliferated space imaging and electronic intelligence, commercial radio geolocation, and fusion software at scale. Not by passive “ambient” physics.
   HIGH
3. KJ-3
   Submarines remain far harder to find than surface ships. The credible threats are expanding acoustic networks in chokepoints and littoral bastions, plus incremental magnetic-detection gains. Regional and short-range, not global.
   MODERATE
4. KJ-4
   Deception and emission control retain significant value, especially against long-range kill chains that need targeting-quality tracks rather than mere detection.
   MODERATE

Caveats

Classified capability (undersea surveillance performance, NRO sensor specifics, Chinese seabed coverage) cannot be assessed from open sources; this briefing grades only what is publicly evidenced. Several requested items were not traced to primary documents in this cycle: FY2025 to FY2027 budget lines for Overmatch, Linchpin, Golden Dome and commercial imagery; Japanese, British and Norwegian arrays; the 2005 Gotland encounter; primary PLA doctrinal texts. Passive-radar figures are mostly manufacturer or partner claims. Chinese research is often reachable only through SCMP summaries, whose headlines overstate the papers beneath them.

Sources

Open sources consulted

Grouped by chapter. Chained citations are declared as such in the text.