The Signal, Issue #005

Three things failed this week, and in each case the failure happened at a layer below where the defence was built. A court order did not stop a spyware operation. A sovereign platform did not survive a single compromised account. A reference list did not survive a plausible-sounding fabrication. The technical and legal architecture held. The human layer did not, and the scientific layer is worse than most practitioners have accounted for.

NSO Group resumed Pegasus operations after a permanent injunction. Meta is back in court.

On June 8, Meta filed a contempt motion against NSO Group in a US federal court, alleging that the Israeli spyware firm had violated a permanent injunction issued in 2025 almost immediately after it was handed down. Meta's threat intelligence team documented at least 23 WhatsApp accounts and 34 groups linked to NSO created between January and early June 2026, the majority used for testing and some for active targeting. The campaign used deceptive links designed to push users off WhatsApp to external sites, a one-click phishing technique consistent with previously documented NSO tradecraft. Targets were identified in Jordan and Lebanon. Meta disrupted the operation after investigating user reports, then moved to court.

The practitioner read here is not primarily about NSO. It is about what a court injunction does and does not constrain. The 2025 ruling barred NSO from targeting WhatsApp users and extracted $168 million in damages across a six-year civil case. NSO is appealing the ruling. In the interval between the injunction and the appeal outcome, the infrastructure kept running. The legal order did not pause the operational posture. For any investigator modelling the threat environment around a high-risk source or subject, the NSO contempt motion is a data point about how persistent spyware actors treat legal constraints: as a variable in the risk calculation, not as a ceiling on operations.

The contempt motion itself is worth reading for the detection methodology Meta describes: user reports triggered the investigation, which then found the test accounts and groups through platform telemetry. The operational chain ran from human reporting to infrastructure discovery. That sequence matters for newsrooms and civil society organisations managing high-risk contacts: if a target's device is compromised and they report suspicious behaviour, the evidentiary trail from that report to the infrastructure is now documented in a court filing. The detection model is user-initiated, not automated. The implication is that targets who do not recognise or report the approach are not in the dataset.

Tchap's architecture held. One socially-engineered account was enough.

On June 7, ANSSI detected a breach of Tchap, the French government's sovereign messaging platform. Tchap was built on the open-source Matrix protocol by DINUM, France's interministerial digital directorate, specifically to keep government communications on state-managed infrastructure rather than foreign platforms. In August 2025, Prime Minister Bayrou mandated its use for all civil servant work communications and banned WhatsApp and Signal for official use. By the time of the breach, Tchap had over 800,000 registered civil servants across all major ministries. The attacker gained access through a socially-engineered account on Tchap's education environment. DINUM identified and blocked the compromised account and launched an investigation.

An unverified attacker using the handle Misère claimed access to around 73,000 accounts, 643,000 messages, 13.5GB of files, and approximately 90 items carrying a Diffusion Restreinte classification marking -- none of which DINUM or ANSSI confirmed. What is confirmed: one account, one social engineering approach, access to public unencrypted rooms. The architecture performed as designed; the failure was not in the platform.

The structural question the Tchap breach puts to any organisation running a mandated platform migration is specific: what changed in the human-layer security posture when the platform changed? A ban on WhatsApp and Signal removes the foreign infrastructure dependency. It does not transfer the operational security practices that experienced users of those platforms had developed over years. A civil servant who understood not to share sensitive material in a WhatsApp group chat may not have the same instinct about a public Tchap room, particularly in the first year of a mandated transition. The investigation is ongoing. Until DINUM and ANSSI publish findings on how the account was compromised and what the attacker could access, treat the unverified claims as indicators rather than confirmed facts, and the confirmed access as the floor rather than the ceiling.

The CITADEL audit found 4,046 fabricated citations. That number is a floor, not a count.

The Topaz et al. correspondence published in The Lancet on May 9 presents CITADEL, the first systematic audit of reference integrity across the biomedical literature at scale. The team scanned 2,471,758 papers and 125,615,773 structured references in the PMC Open Access subset from January 2023 to February 2026. Of those references, 77% carried a PubMed identifier and were verifiable. The remaining 23%, covering websites, books, and grey literature, were excluded. Among the verified references, the pipeline identified 4,046 fabricated citations across 2,810 papers. The fabrication rate rose from approximately four per 10,000 papers in 2023 to 56.9 per 10,000 in early 2026, a 12-fold increase. Review articles had a fabrication rate 57% higher than other paper types. Of the 2,810 affected papers, 98.4% had received no publisher action at the time of the audit.

The 4,046 number is the figure that will travel, and it is the wrong number to anchor on. The authors are explicit about this: the pipeline estimates precision, not recall. Fabricated references that evaded all filters are not counted. The 23% of references excluded for lacking a PMID represent a population where fabrication may be more or less common, and the audit has no estimate for either direction. PMC Open Access is not the full biomedical literature. The early 2026 data covers seven weeks. The confirmed fabrications are what the pipeline caught within its scope. The scope excluded most of the literature.

What this means for practitioners using biomedical literature as a source layer

The fabricated citations in this dataset were not obviously defective. Topically specific, correctly formatted, attributed to real researchers, bearing plausible publication dates. The pipeline caught them because it could cross-check identifiers against four independent databases. A working journalist or investigator reading the citing paper has none of that infrastructure in their workflow. The audit also identified paper mill patterns: the same two authors appearing across 11 papers in a single surgical journal in 2025, with 15 fabricated references covering CRISPR diagnostics, AI-guided nanovaccines, and gut microbiome biomarkers, all sharing a core co-authorship pair. That is a coordinated contamination of the literature, not a noise problem. For investigators using biomedical literature to support claims about health, pharmaceutical, or environmental topics, the workflow implication is immediate: a published, peer-reviewed citation is no longer independently sufficient evidence that the cited study exists and says what the citing paper claims. Verify the primary source directly. That step is not optional.

TEL-002: Trace a mobile number to a named subscriber identity

Both the NSO contempt motion and the Tchap breach reduce to the same operational question: how do you get from a platform, a number, or an account to a named person? The NSO campaign was detected when users reported suspicious links. The Tchap attacker entered through a single account. In both cases, the investigative move that matters is attribution: working from the contact point back to the actor.

TEL-002, published this week, is the Signal & Shadow reference card for tracing a mobile number to a named subscriber identity. It walks the MSISDN parse, HLR status confirmation, pivot to social media and breach records, and the construction of a graded attribution dossier. The card covers what each data source establishes, what it cannot establish, and where the confidence grade shifts.

Available to Signal and Shadow tier subscribers.

The work continues.

Derek

The Signal is the weekly intelligence briefing from Signal & Shadow, an independent forensic investigation and methodology practice.

Signal & Shadow  ·  signalandshadow.io  ·  Issue #005  ·  19 June 2026