AI Wiki

Smart Glasses and Privacy

AI EthicsAI HardwareAI Policy & Regulation
21 min read
Updated
Suggest editHistoryTalk29 citationsFact-checked
RawGraph

Camera-equipped smart glasses pose a privacy problem that phones do not, because the person being recorded usually cannot tell. The industry's answer since Facebook and Luxottica launched Ray-Ban Stories in September 2021 has been a small indicator light, and European regulators began questioning whether it works within days of that launch [1][2]. By mid-2026 the dispute had moved well past the LED: thirteen putative class actions against Meta Platforms and Luxottica of America were consolidated in the Northern District of California [3], the Texas attorney general opened an investigation [4], New York barred camera eyewear from every courthouse in the state [5], and reporting on dormant face-recognition code inside Meta's companion app produced a public argument over whether the feature existed at all [6][7]. Almost none of it has been resolved. As of 2026-08-01 no court had ruled on the merits of any privacy claim about smart glasses, and no manufacturer had shipped face recognition on a consumer pair.

The bystander problem

Recording someone with a phone involves a visible gesture. The device comes up, the framing is obvious, and the subject can object before or during the capture. Glasses remove all three signals. The camera sits where a hinge would be, points wherever the wearer looks, and is available continuously rather than in deliberate bursts. The Irish Data Protection Commission described the asymmetry precisely in 2021: while many devices including smart phones can record third parties, "it is generally the case that the camera or the phone is visible as the device by which recording is happening, thereby putting those captured in the recordings on notice" [1].

That is a consent problem rather than a data-security one, and it falls on people who never bought the product. The wearer accepts terms of service; the bystander is simply in frame. This is why the recording indicator, rather than encryption or on-device processing, became the industry's central privacy artifact, and why so much regulatory attention has landed on a component costing a few cents.

The capture LED in practice

Meta's operative public statement on the indicator is a newsroom post, "Meta's AI Glasses: Your Questions Answered," published 2026-07-07 and updated 2026-07-23 [8]. The post makes three claims worth separating.

Meta's claim (verbatim)Scope and date
"The capture LED has no off switch." Meta adds that while phones and action cameras lack such a light, "ours have had them since day one"All generations, per Meta [8]
"Beginning with our second generation of glasses, the camera is automatically disabled if we detect that the capture LED has been blocked. No photos or videos can be taken until we detect that the light is unblocked"Second generation onward only; Meta does not say whether Ray-Ban Stories or first-generation Ray-Ban Meta ever received it [8]
"We also take legal action against people or businesses that sell services designed for tampering with the capture LED, both on and off our own platforms"Stated as standing policy; Meta names no defendant and cites no case [8]

Meta also says it "conducted significant testing to find the right level of brightness so they are visible even during the day" [8]. In July 2026 it extended the safeguard, saying it was "updating the glasses to disable the camera if they detect the LED was physically tampered with or destroyed," and claiming that "no other kind of camera has done this" [8].

The Texas attorney general's office disputes part of this picture. Its announcement of the investigation says the indicator "is easily hidden" and adds: "Furthermore, this LED indicator is not active during the smart glasses' 'always enabled' mode," the mode in which, on the office's reading of Meta's privacy policy, the device constantly processes video for Meta AI [27]. Meta has not conceded that characterisation, and the point has not been tested anywhere.

The safeguard has also been defeated in public. On 2026-07-31 Engadget reported that a notched plastic-and-sticker cover costing roughly two dollars beat the blocked-LED check: the notches admit enough ambient light for the sensor to read the LED as unobstructed, while an opaque layer hides the glow from anyone nearby, and because the stickers leave the hardware intact the newer physical-tampering check has nothing to find [9]. Meta told Engadget the same day that such products "violate our policies, so we're exploring ways to strengthen our ability to detect such tampering and disable the camera" [9]. That is an unfinished arms race, not a solved problem.

Regulatory scrutiny of the indicator

Ray-Ban Stories went on sale in Ireland and Italy in September 2021, and both countries' regulators reacted within a week. The Italian Garante asked the Irish DPC on 2021-09-10 to press Facebook for answers before Italian sales began, seeking in particular "the legal basis for Facebook to process the personal data; the measures in place to safeguard bystanders, especially children; any mechanisms to anonymise the collected information; the features of the voice assistant used to operate the glasses" [2].

On 2021-09-17 the DPC issued a joint statement with the Garante. Its criticism was narrow and evidentiary rather than rhetorical: "there is a very small indicator light that comes on when recording is occurring. It has not been demonstrated to the DPC and Garante that comprehensive testing in the field was done by Facebook or Ray-Ban to ensure the indicator LED light is an effective means of giving notice" [1]. The two authorities called on Facebook Ireland "to confirm and demonstrate that the LED indicator light is effective for its purpose and to run an information campaign to alert the public" [1]. Neither regulator issued a fine or a formal finding of infringement, and neither has published a decision on the glasses since.

Human review and the annotation chain

The reporting that most of the 2026 litigation rests on is a joint investigation by the Swedish newspapers Svenska Dagbladet and Goteborgs-Posten, published 2026-02-27 [10]. Reporters interviewed data annotators at Sama, a Meta subcontractor in Nairobi, Kenya, who described reviewing captured audio and video from the glasses. One worker said: "In some videos you can see someone going to the toilet, or getting undressed. I don't think they know, because if they knew they wouldn't be recording" [10].

The reporting also addressed Meta's automatic face blurring. Former Meta employees said faces in annotation data are blurred automatically, but annotators in Kenya said the anonymisation does not always work as intended. Asked how, one former employee said: "The algorithms sometimes miss. Especially in difficult lighting conditions, certain faces and bodies become visible" [10]. Meta's London spokesperson Joyce Omope replied after two months without answering the specific questions, referring the papers to Meta's AI terms of use and privacy policy; Sama did not respond [10]. None of this has been established anywhere but in the reporting itself. Meta has not confirmed the annotators' accounts, no regulator has made a finding about them, and no court has weighed them. Human review of training material is ordinary industry practice, since data labeling and human-in-the-loop review underpin most large models. That is part of the plaintiffs' point: they say buyers were not told it applied here.

Face recognition

Face recognition would turn a wearable camera into an identification device. No consumer glasses product ships it, but the components are all commercially available, and the clearest demonstration of that came from two Harvard students, AnhPhu Nguyen and Caine Ardayfio, who posted a video on 2024-09-30 of a system they called I-XRAY [11][12]. Their own write-up states the purpose plainly: "The purpose of building this tool is not for misuse, and we are not releasing it," and says the goal was "to demonstrate the current capabilities of smart glasses, face search engines, LLMs, and public databases, raising awareness that extracting someone's home address and other personal details from just their face on the street is possible today" [11]. They declined to name the five technologies they combined, but their own opt-out guide identifies the categories: reverse face search (PimEyes, Facecheck.id) and people-search databases such as FastPeopleSearch [11]. A large language model stitched the results into an identity, after which, in their words, "a FastPeopleSearch lookup can identify the person's home address, phone number, and their relatives" [11].

Two details matter. The pipeline streamed video from the glasses to Instagram Live for a separate program to monitor, and Engadget said "it can take just a couple of minutes for this process to play out" [12]; headlines describing instant identification overstate what was shown. The students chose Ray-Ban Meta glasses because "they look almost indistinguishable from regular glasses," and Nguyen framed the risk concretely: "Some dude could just find some girl's home address on the train and just follow them home" [12]. I-XRAY was a student demonstration, never released and never sold.

Meta's own effort surfaced in February 2026, when reporting by the New York Times on internal documents disclosed a face-recognition feature for the glasses under the internal name "Name Tag" [27], along with an internal memo saying the company would "launch during a dynamic political environment where many civil society groups that we would expect to attack us would have their resources focused on other concerns" [6][13]. On 2026-04-13 seventy-seven organisations, among them the American Civil Liberties Union, the Electronic Frontier Foundation, EPIC and Access Now, urged Meta to "immediately halt and publicly disavow its plans to deploy facial recognition features on its Ray-Ban and Oakley glasses" [28][13].

The code itself was reported on 2026-06-04, when WIRED said it had found a dormant face-recognition pipeline inside the Meta AI companion app, which had been installed on more than 50 million phones [7]. EFF's Threat Lab said it confirmed the code was present through static analysis of the application, and described faceprints stored as "a series of 2,048 numbers uniquely representing the positioning of a person's facial features," which the feature would compare against every new face coming into the wearer's view [6]. Meta shipped an app update on 2026-06-05, one day later, that removed nearly all of it [6][7].

Meta's public account has been inconsistent, and the inconsistency is itself the story. Before publication, communications vice-president Andy Stone said "the feature does not exist"; chief technology officer Andrew Bosworth called the reporting "incredibly misleading" and "absolutely dishonest"; after publication Stone described the feature as "purely exploratory" with no final decision made [7]. Bosworth later described it on a podcast as recall of people the wearer has already met rather than identification of strangers, a materially narrower capability than I-XRAY demonstrated [29]. Meta declined to say why the code was removed or whether it would return [7].

Nothing here has shipped. As of 2026-08-01 Meta had not released facial recognition on any pair of its glasses [7].

Litigation

Thirteen putative class actions filed in March 2026 were consolidated in the United States District Court for the Northern District of California, San Francisco Division, under the caption In re Meta AI Glasses Litigation, master file 3:26-cv-01897-EMC, before Judge Edward M. Chen [3]. The consolidation stipulation describes them as arising "out of allegedly false advertising and/or privacy-related issues related to purchase and/or use of the Meta AI Glasses," each naming at least Meta Platforms and Luxottica of America as defendants; seven also name Samasource Impact Sourcing Inc., doing business as Sama [14].

DateEvent
2026-03-04Bartone v. Meta Platforms filed, the first of the thirteen [15]
2026-03-19Boylett v. Meta Platforms filed, 3:26-cv-02409 [16]
2026-04-22Judge Chen grants the stipulation consolidating the actions and directs the clerk to administratively close the member cases, Boylett among them [14]
2026-07-19Interim co-lead counsel appointed: Thomas E. Loeser, Ryan J. Clarkson and Tina Wolfson, as individuals rather than their firms [17]
circa 2026-09-02Consolidated complaint due, 45 calendar days after the appointment [14]

The first-filed complaint, brought by Gina Bartone and Mateo Canu, alleges that Meta marketed the glasses as "designed for privacy, controlled by you" while routing captured audio and video to human annotators at Sama in Nairobi, and that the advertised face anonymisation does not reliably work [15]. It pleads consumer-protection and contract theories rather than privacy torts: California's Unfair Competition, False Advertising and Consumer Legal Remedies Acts, the New Jersey Consumer Fraud Act, fraud, negligent misrepresentation, breach of contract, breach of implied warranty and unjust enrichment [15].

Boylett is the wiretap case. Filed by Jeffrey Boylett, it pleads four counts: the California Invasion of Privacy Act at Penal Code sections 631 and 632, the Federal Wiretap Act at 18 U.S.C. 2510 and following, intrusion upon seclusion, and unjust enrichment [16].

Meta's public response came through spokesperson Christopher Sgro, who did not address the litigation itself: "Unless users choose to share media they've captured with Meta or others, that media stays on the user's device. When people share content with Meta AI, we sometimes use contractors to review this data for the purpose of improving people's experience, as many other companies do. We take steps to filter this data to protect people's privacy and to help prevent identifying information from being reviewed" [18].

Every claim above is an allegation. As of 2026-08-01 the consolidated complaint had not been filed, no defendant had answered or moved to dismiss, no class had been certified, and no court had ruled on the merits [3]. The most recent docket activity was an amended protocol for common-benefit work filed 2026-07-28 [3].

What has not been filed

A search of federal docket data found no Illinois Biometric Information Privacy Act case, no Texas Capture or Use of Biometric Identifier Act case, and no product-liability case about smart glasses. That is a "no filed case has been identified" finding rather than proof that none exists, because coverage of state trial courts in the available docket corpora is thin. One parallel state action does appear on the federal docket: a notice of pendency of another action under Civil Local Rule 3-13, filed 2026-04-27, flagged Burman v. Meta Platforms, Inc., Los Angeles Superior Court No. 26STCV07570 [3]. That entry gives the caption and nothing else, so what Burman pleads is not known from the federal record.

The Texas matter is an investigation, not a suit: on 2026-05-20 attorney general Ken Paxton announced that his office had issued a civil investigative demand to Meta, saying "Meta's glasses raise serious concerns, and my office will thoroughly investigate these devices to ensure that no individual is being unlawfully recorded, tracked, or subjected to the unauthorized collection of their data" [27]. The same release says Paxton had secured a $1.4 billion settlement with Meta in July 2024 over earlier facial-recognition claims [27]. A civil investigative demand is a compulsory request for documents and answers; it is not a finding, and no Texas enforcement action about the glasses has been filed.

Regulation

The EU AI Act, Regulation (EU) 2024/1689, is the most commonly cited instrument here and the most commonly misdescribed. Its Article 5(1)(h) prohibition on "real-time" remote biometric identification in publicly accessible spaces applies only "for the purposes of law enforcement" [19]. It does not reach a consumer wearing a camera. The provision that actually bites on a consumer face-search product is Article 5(1)(e), which bans AI systems "that create or expand facial recognition databases through the untargeted scraping of facial images from the internet or CCTV footage," because that is how the reference database behind such a product gets built [19]. Article 5(1)(f) prohibits emotion inference specifically "in the areas of workplace and education institutions" [19].

ProvisionRelevance to camera glassesApplies from
Art. 5(1)(e)Bans building or expanding face-recognition databases by untargeted scraping2025-02-02 [19]
Art. 5(1)(f)Bans emotion inference in workplaces and schools2025-02-02 [19]
Art. 5(1)(h)Real-time remote biometric ID, law enforcement only2025-02-02 [19]
Art. 50(3)Deployers of emotion recognition or biometric categorisation must inform those exposed2026-08-02 [19]

Under Article 113 the Act applies generally from 2 August 2026, with Chapters I and II, which contain the Article 5 prohibitions, applying from 2 February 2025 [19]. Regulation (EU) 2026/1744, the Digital Omnibus on AI, was adopted 8 July 2026 and entered into force 27 July 2026 [20]. It defers the high-risk obligations in Chapter III to 2 December 2027 and 2 August 2028, but leaves the biometric prohibitions and the Article 50 transparency duties on their existing dates [20].

For bystanders, the operative European question is the GDPR household exemption. Article 2(2)(c) excludes processing "by a natural person in the course of a purely personal or household activity," and Recital 18 adds that this requires "no connection to a professional or commercial activity" while confirming that the Regulation still "applies to controllers or processors which provide the means for processing" [21]. The closest judicial guidance is the Court of Justice's 2014 judgment in Case C-212/13, Rynes, holding that a home security camera which "also monitors a public space, does not amount to the processing of data in the course of a purely personal or household activity" [22]. Two caveats: Rynes construed the second indent of Article 3(2) of Directive 95/46/EC, the earlier Data Protection Directive, and not the GDPR, and it concerned fixed CCTV. No Court of Justice case has applied it to a wearable camera, so it is the governing analogy rather than a holding about glasses.

In the United States, the statutes actually reached so far are those pleaded in the consolidated litigation [15][16]. No state has enacted a smart-glasses-specific recording statute, though bills have been introduced.

Rules in practice

Institutions have moved faster than legislatures. On 2026-07-20 the New York State Office of Court Administration began prohibiting smart glasses containing a camera, microphone or computer in state, county, town and village courts, the first US state to do so [5]. The New York State Bar Association reported that a memorandum dated 2026-07-01 gave as the rationale that the ban ensures no one can be recorded surreptitiously, which is a violation of the New York State Civil Rights Law; that wording is the association's summary, because the memorandum itself is not publicly readable [5]. The rule binds attorneys, litigants, witnesses, family members, court employees and the public alike, glasses brought into a court facility should be turned over to uniformed court officers for safekeeping, and wearers of prescription smart lenses are told to bring a conventional pair [5]. The concern had been raised earlier in 2026 when members of Mark Zuckerberg's security detail wore smart glasses during his trial in California and a Superior Court judge warned them they would be held in contempt if they recorded [5].

Similar restrictions have been reported at cinemas, casinos, examination halls, secure facilities and security conferences, most of them extending rules originally written for Google Glass a decade earlier. They are institutional policies rather than law, they are rarely published as citable documents, and no comprehensive survey of them exists.

Security research

The published vulnerability record for smart glasses is thin. Two CVE records are on point.

IdentifierProductDescriptionSeverity
CVE-2021-24046Ray-Ban Stories device software before 2107460.6810.0"A logic flaw in Ray-Ban Stories device software allowed some parameters like video capture duration limit to be modified through the Facebook View application" [23]None published
CVE-2026-4217XREAL Nebula App 3.2.0 and 3.2.1 on AndroidUnprotected storage of credentials in a cloud-storage plugin; local access only, high attack complexity; the vendor "did not respond in any way" to the disclosure [24]CVSS v4.0 2.0, LOW

Both were verified against the CVE Program's own records [23][24]. Keyword sweeps of the National Vulnerability Database index and the GitHub Security Advisory database turned up no other smart-glasses CVE, which is a "none identified" result rather than proof that none exists. Neither of the two is a bystander-privacy flaw; the record contains nothing resembling remote camera activation.

Platform responses in 2026

In late July 2026 Instagram head Adam Mosseri said in an Instagram Story that the platform would remove videos filmed on Meta's glasses to harass people: "If you're posting content that is taking advantage of people and harassing them, like a lot of these pickup line kind of videos, then we're going to take the content down. We don't want people to be surreptitiously taking videos of other people and harassing them and then posting them on our platform" [25]. At least two accounts, each with more than a million followers, were deactivated [25]. This was an informal enforcement statement, not a documented policy change: no Community Guidelines revision or help-centre page naming smart glasses has been identified, and Meta declined to explain how the rule is enforced or what counts as a violation [25].

Calls for outright bans have come mostly from outside government. In late July 2026 Macquarie University's news service published comments by Professor Dali Kaafar, executive director of its Cyber Security Hub, arguing that Australia should ban or restrict the glasses "except in genuine use cases such as disability," and criticising the marketing of Meta's Kylie Jenner collaboration [26]. Kaafar is an academic, not an Australian government official, despite the "cyber chief" framing used in syndicated headlines, and no Australian government action has been identified.

See also

References

  1. ^"Data Protection Commission statement concerning Facebook View (glasses)." Data Protection Commission (Ireland), 2021-09-17. dataprotection.ie/...cerning-facebook-view-glasses
  2. ^"Smart glasses: Garante Privacy chiede informazioni a Facebook / Italian SA seeks information from Facebook." Garante per la protezione dei dati personali, doc-web 9698698, 2021-09-10. garanteprivacy.it/...9698698
  3. ^"In re Meta AI Glasses Litigation, No. 3:26-cv-01897 (N.D. Cal.), docket." CourtListener / Free Law Project, accessed 2026-08-01. courtlistener.com/...re-meta-ai-glasses-litigation
  4. ^Kimery, Anthony. "Texas AG opens investigation into Meta glasses over privacy, biometric concerns." Biometric Update, 2026-05-21. biometricupdate.com/...-privacy-biometric-concerns
  5. ^Andrus, Jennifer. "New York State Court System Bans Wearing Smart Glasses in Court Buildings." New York State Bar Association, 2026-07-20. nysba.org/...ring-smart-glasses-in-court-buildings
  6. ^Quintin, Cooper. "Move Fast, Surveil Things." Electronic Frontier Foundation, 2026-06-04 (updated 2026-06-08). eff.org/...move-fast-surveil-things
  7. ^Constantin, Ana Maria. "Meta pulled facial recognition code from its smart glasses app one day after WIRED found it, then denied the timing was related." The Next Web, 2026-06-09. thenextweb.com/...ecognition-smart-glasses-removed
  8. ^"Meta's AI Glasses: Your Questions Answered." Meta Newsroom, 2026-07-07, updated 2026-07-23. about.fb.com/...ai-glasses-your-questions-answered
  9. ^"A $2 Sticker Let Me Bypass The Meta Glasses' Anti-Creep Feature." Engadget, 2026-07-31. engadget.com/...he-meta-glasses-anti-creep-feature
  10. ^Lepapa, Naipanoi; Abdigadir, Ahmed; Lindblom, Julia; Norman, Erik. "Meta's AI Smart Glasses and Data Privacy Concerns: Workers Say 'We See Everything'." Svenska Dagbladet / Goteborgs-Posten, 2026-02-27. svd.se/...y-concerns-workers-say-we-see-everything
  11. ^Nguyen, AnhPhu; Ardayfio, Caine. "I-XRAY" (project document). 2024-09-30. docs.google.com/...edit
  12. ^Holt, Kris. "Students used Meta's smart glasses to automatically dox strangers via Instagram streams." Engadget, 2024-10-02. engadget.com/...rs-via-instagram-streams-170228496
  13. ^Borak, Masha. "Meta urged to halt facial recognition glasses rollout, disclose law enforcement ties." Biometric Update, 2026-04-14. biometricupdate.com/...sclose-law-enforcement-ties
  14. ^"Stipulation and Order Re Consolidation of Actions and Setting Process for Appointment of Interim Class Counsel." In re Meta AI Glasses Litigation, Dkt. 57, N.D. Cal., 2026-04-22. storage.courtlistener.com/....cand.465369.57.0.pdf
  15. ^"Bartone et al. v. Meta Platforms, Inc. and Luxottica of America, Inc., complaint." N.D. Cal. No. 3:26-cv-01897, filed 2026-03-04. classaction.org/...bartone-v-meta.pdf
  16. ^"Boylett v. Meta Platforms, Inc. et al., complaint." N.D. Cal. No. 3:26-cv-02409, filed 2026-03-19. storage.courtlistener.com/...s.cand.466198.1.0.pdf
  17. ^"Order Granting Motion to Appoint Cotchett Slate; and Denying Motion to Appoint Lowey Slate." In re Meta AI Glasses Litigation, Dkt. 88, N.D. Cal., 2026-07-19. storage.courtlistener.com/....cand.465369.88.0.pdf
  18. ^Hayden, Scott. "Meta Faces Lawsuit Claiming Ray-Ban Smart Glasses Sent Private Footage to Overseas Reviewers." Road to VR, 2026-03-10. roadtovr.com/...suit-ray-ban-smart-glasses-privacy
  19. ^"Regulation (EU) 2024/1689 (Artificial Intelligence Act)." Official Journal of the European Union, 2024-07-12. eur-lex.europa.eu/...HTML
  20. ^"Regulation (EU) 2026/1744 of 8 July 2026 amending Regulations (EU) 2024/1689, (EU) 2018/1139 and (EU) 2023/1230 (Digital Omnibus on AI)." Official Journal of the European Union, 2026-07-24. eur-lex.europa.eu/...TXT
  21. ^"Regulation (EU) 2016/679 (General Data Protection Regulation)." Official Journal of the European Union, 2016-05-04. eur-lex.europa.eu/...HTML
  22. ^"Judgment of the Court (Fourth Chamber), Case C-212/13, Rynes v Urad pro ochranu osobnich udaju." Court of Justice of the European Union, 2014-12-11. eur-lex.europa.eu/...HTML
  23. ^"CVE-2021-24046." CVE Program record, published 2022-01-14. cveawg.mitre.org/...CVE-2021-24046
  24. ^"CVE-2026-4217." CVE Program record, published 2026-03-16. cveawg.mitre.org/...CVE-2026-4217
  25. ^Constantin, Ana Maria. "Meta is banning the 'pervert glasses' creeps from Instagram." The Next Web, 2026-07-24. thenextweb.com/...s-harassment-ban-pervert-glasses
  26. ^"Cyber Chief Seeks Ban on Meta's Kylie Jenner AI Glasses." Mirage News, republishing Macquarie University / The Lighthouse, 2026-07-31 AEST. miragenews.com/...on-metas-kylie-jenner-ai-1719559
  27. ^"Attorney General Ken Paxton Launches Investigation Into Meta Glasses to Protect Texans' Privacy From Unlawful Monitoring and Collection of Facial Data." Office of the Attorney General of Texas, press release, 2026-05-20. texasattorneygeneral.gov/...exans-privacy-unlawful
  28. ^"Sign-On Letter to Meta Regarding Facial Recognition Features on Meta Glasses," 2026-04-13, hosted by the ACLU of Massachusetts. aclum.org/...Features-on-Meta-Glasses-April-14.pdf
  29. ^Harrison Dupre, Maggie. "Top Meta Exec Describes Controversial Facial Recognition Feature in Detail After the Company Claimed It Didn't Exist." Futurism, 2026-07-18. futurism.com/...scribes-facial-recognition-feature

Improve this article

Add missing citations, update stale details, or suggest a clearer explanation. Every suggestion is reviewed for sourcing before it goes live.

v1 · 4,204 words · full history

Fact-checks are independent of edits: a reviewer re-verifies the article against its sources and stamps the date. How we verify

Reviewer note: Independent adversarial fact-check, 2026-08-02: every load-bearing number, date, price, spec and quotation re-checked against primary sources by a second agent; citation markers audited against their references.

Suggest edit