655 points by rzk 5 days ago | 198 comments on HN
| Moderate positive
Contested
Human Rights · v3.7· 2026-02-26 04:23:14 0
Summary Privacy & Surveillance Champions
This investigative report exposes an identity surveillance infrastructure built by Persona and OpenAI with integration into U.S. federal agencies including ICE, documenting 27 months of continuous operation screening millions of users monthly against watchlists using facial recognition and intelligence database matching. The content champions human rights through detailed technical investigation protected under First Amendment, ECHR Article 10, and international human rights frameworks, with explicit legal disclaimers establishing good-faith journalism and transparent methodology that enables public scrutiny of surveillance systems threatening privacy, due process, and freedom of expression.
It seems like at every technological step, we're sold the dream and delivered the meme. We always end up with the worst possible combination of players, ideas and outcomes; with the promise of what the said technology delivers in terms of additional freedom or free time never realised. How many more broken social contracts can society endure before it crumbles?
Please note that Persona primarily operates as a "service provider" or "processor" for its customers. We act as a "business" or "controller" only for specific services, such as identity verification for LinkedIn, FoxCorp, and Reusable Persona. To learn more about how Persona manages your personal data, please refer to our privacy notices, which can be accessed through the following link: https://withpersona.com/legal/privacy-notices
If you wish to exercise your privacy rights related to services where Persona is a "service provider" or "processor," please contact the entity using our service, as they are the "controller" of the data. We will assist the relevant customer to fulfill your data subject rights, but we do not handle such requests directly on their behalf.
For any privacy rights request related to services where Persona acts as a "business" or "controller," including identity verification for LinkedIn, FoxCorp, Reusable Persona, and personal data related to our sales, marketing activities, or website browsing on withpersona.com, please use our Data Subject Request (DSAR) available at the following link: https://withpersona.com/dsar
For all other inquiries, we will respond as soon as possible.
"what is Fivecast ONYX? an AI-powered surveillance platform purchased by ICE for $4.2 million and CBP for additional license costs. according to Fivecast’s own documentation and EFF’s reporting, they do automated collection of multimedia data from social media and dark web, build “digital footprints” from biographical data, tracks shifts in sentiment and emotion, assigns risk scores, searches across 300+ platforms and 28+ billion data points, identifies people with “violent tendencies”"
Glad to know that my tinfoil hat wasn't too tight when social media came to be and this obvious use was predicted. How quickly will not having social media accounts become a crime?
Governments in Europe should be seriously scrutinising this with the background conversation of departing American tech going on. Discord users globally were being coerced into handing over their ID to this American surveillance tech. Are we just going to let this go on?
> OpenAI’s disclosures reference biometric data stored “up to a year.” the source > code shows face list retention capped at 3 years. government IDs retained
> “permanently” per Persona’s practices. which is it?
I keep saying this. This is the playbook -- everything is moving to standardize Sam Altman's biometric authentication cryptocurrency company to use internet services. This has been a slow moving strategy for /years/ and every new step over that period only get closer, not further from this goal.
"We weren’t hacked" is doing PR triage for "we exposed sensitive internal implementation details." Spy company semantics are always incredible. The house didn’t burn down, it just leaked gas.
calling data sovereignty laws a cybersecurity risk in the same week that Persona had 2500 files exposed on a government endpoint is an interesting choice of timing.
2. Request deletion. The verification is done. LinkedIn already has the result. There is no reason for Persona to keep your passport scan and facial geometry on their servers. Ask them to delete it.
3. Contact their DPO. [email protected] — that’s their Data Protection Officer. If you want to object to them using your documents as AI training data under “legitimate interests,” this is where you do it.
4. Think twice before verifying. That blue badge might not be worth what you’re trading for it. A checkmark is cosmetic. Biometric data is forever.
Organize in your country and advocate for data deletion jubilees, organize in your country to champion new taxes against US digital services, organize in your country to advocate for homegrown solutions over US tech.
If you aren't actively organizing you aren't going to accomplish anything.
Remember that people power trumps monetary power, but you have to commit for people power to work.
From my understanding, we are pretty close to a Dystopian world where all elites of a certain group collaborate to run a Super Leviathan. We still gotta choose our flavors, which may not be feasible in maybe 5-10 years when those leviathans clash into each other.
It's "socializing the losses and privatizing the gains"… but now alarmingly supercharged well beyond purely financial realms, and into really basic and fundamental matters of individual physical autonomy and liberty.
I get the KYC concerns for API access, but I'm sortof baffled at why they'd need all of the AML stuff, given that they're not payment processors/financial institutions.
Or does Persona provide that by default? Don't know much about their service...
Many tech execs operate under the thesis that china & the democratic party are existential threats that warrant a surveillance/military/police ramp up. Meanwhile, many tech employees are credulous and frequently adopt self-serving geopolitical narratives. The current macro trends don't help (huge defense budgets, bad labor market power, China is in fact more powerful)
According to Persona's damage control article[0], the subdomain had "onyx" in its name because that's the internal code name for the project, and it's named after the pokémon Onyx. No connection to Fivecast ONYX.
This is the same complete bullshit trying to remove oneself from political donation emails. "Oh, okay, we will remove you from that one." Days later it's a "different campaign." Sometimes it's the exact same people from weeks ago who have just renamed their campaign and started sending again.
We need far stronger laws for all of it, which will never happen because the rot and corruption has fully metastasized.
> How many more broken social contracts can society endure before it crumbles?
Having any kind of agency in those things would be a start.
If <FAANG bigcorp of your choice> announces with great fanfare "We're building this totally awesome new technology that will make everything better! And the best thing? You won't have to do anything, we will auto-update all your devices/accounts/etc with it for free! Trust us!", then whether you personally believe their enthusiastic predictions or not doesn't really matter a lot - you will get it anyway, unless you spend a lot of energy to deliberately avoid the new technology.
That does not match the very similar reply I got as a California resident asserting my rights under California's "Right to Know" Act , regarding LinkedIn profile data and related
You act like the governments of Europe weren’t the reason Discord decided they needed to get government issued identity information from European users…
Central human rights engagement. Content is protected journalism and security research investigating government surveillance. Authors explicitly invoke First Amendment, ECHR Art. 10, CFAA safe harbor, California Shield Law, and GDPR Art. 85 as legal basis for publication. Document demonstrates commitment to freedom of expression through detailed investigation, transparent methodology, and explicit legal analysis.
FW Ratio: 50%
Observable Facts
Legal notice states: 'this is protected journalism and security research under the First Amendment, ECHR Art. 10, CFAA safe harbor (DOJ Policy 2022), California Shield Law, GDPR Art. 85, and Israeli Basic Law: Human Dignity and Liberty.'
Document includes explicit threat of additional publication if suppression is attempted: 'any attempt to suppress or retaliate against this publication - legal threats, DMCA abuse, employment interference, physical intimidation, or extrajudicial action - will be treated as confirmation of its findings and will trigger additional distribution.'
Authors state findings distributed across multiple jurisdictions and third-party archives before publication.
Methodology documented as passive recon using only public sources (Shodan, CT logs, DNS, HTTP headers, unauthenticated files), with explicit legal citations to Van Buren v. United States and hiQ Labs v. LinkedIn.
Inferences
Authors frame investigation as protected expression invoking multiple international human rights frameworks.
Pre-distribution strategy demonstrates commitment to preventing censorship through geographic and institutional redundancy.
Threat of escalating distribution if suppressed is rhetorical commitment to freedom of expression principle.
Detailed methodology documentation supports claim of good-faith journalism rather than unlawful access.
Core focus of investigation. Content documents systematic violation of privacy: facial recognition, biometric capture, identity dossier collection, continuous re-screening, and integration with government watchlist systems. All without informed consent or meaningful privacy protections. Document states 53 megabytes of unprotected source code was publicly exposed.
FW Ratio: 57%
Observable Facts
Persona API documentation shows complete identity dossier returned including full name, DOB, nationality, address, government document details, front/back ID photos, selfie, and video of identity capture.
System screens millions of users monthly behind the scenes in seconds with customizable facial recognition filters.
Source code exposed on public server included all TypeScript codebase, permissions, API endpoints, compliance rules, screening algorithms, and intelligence integration.
Document reveals continuous re-screening every few weeks without user notification or consent.
Inferences
Collection of complete identity dossier including biometrics without explicit informed consent violates privacy right.
Behind-the-scenes screening without user awareness demonstrates privacy violation intent.
Public exposure of source code indicates negligent privacy infrastructure and lack of basic security practices.
Content advocates for human dignity and universal rights through exposure of mass surveillance infrastructure. Frames identity verification as a violation of fundamental freedoms and explicitly invokes Article 10 (ECHR), First Amendment, and human dignity principles.
FW Ratio: 50%
Observable Facts
Document explicitly cites ECHR Art. 10, First Amendment, CFAA safe harbor, California Shield Law, GDPR Art. 85, and Israeli Basic Law: Human Dignity and Liberty in legal notice.
Content describes surveillance system screening millions of users monthly for watchlist matches and facial recognition flags.
Authors state findings distributed across multiple jurisdictions and third-party archives before publication.
Inferences
Legal citations demonstrate alignment with international human rights frameworks protecting expression and dignity.
Cross-jurisdictional distribution strategy reflects commitment to preventing suppression of information about human rights violations.
Framing surveillance as violation of dignity invokes foundational UDHR principles.
Content directly exposes arbitrary arrest and detention infrastructure. Surveillance system screens users against watchlists without due process; integration with ICE (Immigration and Customs Enforcement) creates risk of arbitrary detention based on algorithmic matching rather than individual assessment.
FW Ratio: 50%
Observable Facts
Document reveals dedicated subdomain 'onyx.withpersona-gov.com' matching ICE's Fivecast ONYX AI surveillance tool with separate deployment.
System screens users against 14 categories of adverse media and creates watchlist tags without judicial process.
Facial recognition similarity scores used to flag users for further investigation without human review threshold.
Inferences
ICE integration indicates surveillance output may inform immigration detention decisions.
Algorithmic watchlist matching without due process creates arbitrary detention risk.
Automated re-screening suggests continuous monitoring for detention triggers without individual consent.
Content exposes systematic interference with property rights through biometric data capture and use. Identity documents, selfies, and videos are collected, stored, and integrated with government surveillance systems without ownership or control by data subjects. Users cannot delete, challenge, or prevent use of their biometric property.
FW Ratio: 50%
Observable Facts
Persona API returns URLs to front/back ID photos, selfies, and identity capture videos without indicating user data ownership or control.
System stores complete biometric profile including facial recognition scores and watchlist similarity matches.
Integration with government platforms and third-party vendors (Sentry, Amplitude, Datadog) extends use of biometric data beyond original purpose without user consent.
Inferences
Users cannot control collection or use of their biometric identity property.
Third-party vendor integration means biometric data is shared with analytics, error tracking, and monitoring services.
Continuous re-screening using biometric data demonstrates ongoing use of property without ongoing consent.
Content addresses freedom of peaceful assembly and association by exposing surveillance infrastructure that enables government to monitor and track users. Watchlist screening and ICE integration create risk that peaceful assembly and association will be chilled by awareness of surveillance.
FW Ratio: 50%
Observable Facts
System screens users against intelligence codenames and 14 categories of adverse media without defining what conduct triggers watchlist.
Continuous re-screening suggests ongoing monitoring for changes in status without user knowledge.
Integration with government platforms and ICE suggests potential law enforcement use against groups engaged in protected association.
Inferences
Vague watchlist criteria create chilling effect on association and assembly.
Continuous monitoring suggests awareness that individuals may be tracked through verification system.
Government platform integration indicates surveillance output may be used against protected groups.
Content advocates for social and international order in which UDHR rights can be fully realized. Document exposes how surveillance infrastructure violates multiple UDHR rights (privacy, expression, assembly, due process). Authors frame investigation as assertion that rights must be protected through transparency and accountability.
FW Ratio: 50%
Observable Facts
Document explicitly invokes UDHR framework: 'the authors are not affiliated with any government, intelligence service, or competitor of any entity named herein.'
Legal notice cites multiple international human rights frameworks (ECHR, CFAA, California Shield Law, GDPR, Israeli Basic Law).
Cross-jurisdictional distribution strategy designed to prevent any single government from suppressing findings.
Inferences
International legal citations demonstrate commitment to universal human rights standards.
Multi-jurisdictional distribution reflects understanding that rights protection requires international cooperation.
Exposure of government surveillance integrated across multiple agencies supports principle of international accountability.
Content explicitly prohibits interpretation of publication as justification for state suppression or elimination of rights. Legal notice warns against retaliation attempts and frames additional distribution as automatic response to suppression, asserting that publication cannot be used to justify removal of authors' rights.
FW Ratio: 57%
Observable Facts
Document states: 'any attempt to suppress or retaliate against this publication - legal threats, DMCA abuse, employment interference, physical intimidation, or extrajudicial action - will be treated as confirmation of its findings and will trigger additional distribution.'
Authors establish that 'killing the messenger does not kill the message' and have pre-positioned evidence with third parties.
Security notice explicitly rejects interpretation of investigation as cause for author harm: 'all authors of this document are in good health, of sound mind, and have no plans to hurt themselves, disappear, or die unexpectedly.'
DCP notes mission to expose surveillance and publish security research in public interest with explicit commitment to First Amendment and CFAA protections.
Inferences
Dead-man's-switch arrangement demonstrates authors' commitment to preventing suppression through protective infrastructure.
Explicit rejection of self-harm framing asserts that investigative journalism cannot justify state violence.
Threat of escalating distribution if suppressed serves as deterrent against interpretation of investigation as cause for rights removal.
Content directly addresses equality and dignity principles by exposing how surveillance infrastructure creates unequal power dynamics. Demonstrates how identity verification processes are weaponized to categorize and control populations based on watchlist status.
FW Ratio: 50%
Observable Facts
Article describes algorithm assigning similarity scores to users and screening them against watchlists, creating system of unequal treatment.
Source code reveals 14 categories of adverse media screening and intelligence codename tagging, showing systematic categorization of users.
Content published without registration or paywall barriers.
Inferences
Exposure of categorization systems demonstrates how surveillance violates equal dignity principle.
Open publication model supports principle that all humans have equal right to understand systems affecting them.
Watchlist screening creates de facto classification of humans into suspect/non-suspect categories.
Content centers on right to life and security of person by exposing surveillance infrastructure that subjects users to ongoing monitoring, watchlist matching, and intelligence codename tagging without consent or transparency. Includes security notice protecting authors from retaliation.
FW Ratio: 50%
Observable Facts
Authors state findings were distributed across multiple jurisdictions and third-party archives before publication as security measure.
Legal notice includes: 'all authors of this document are in good health, of sound mind, and have no plans to hurt themselves, disappear, or die unexpectedly' with dead-man's-switch arrangements.
Document threatens retaliation if authors are harmed, indicating credible threat environment.
Inferences
Authors' security precautions indicate reasonable belief that publication exposes them to retaliation risk.
Distribution strategy designed to preserve information even if authors are harmed, protecting their security of person.
Threats of dead-man's-switch activation serve deterrent function against violence.
Content addresses asylum and refuge implicitly through ICE integration. Document reveals surveillance system designed to screen asylum seekers and immigrants against watchlists, creating risk that identity verification becomes tool for immigration enforcement rather than access enabler.
FW Ratio: 50%
Observable Facts
Document reveals integration with ICE's ONYX surveillance tool through onyx.withpersona-gov.com subdomain.
System screens users against multiple intelligence databases and adverse media categories.
Government platform serves federal agencies with FedRAMP authorization, indicating immigration enforcement use.
Inferences
Integration with ICE suggests asylum seekers and immigrants are subject to heightened surveillance through identity verification.
Facial recognition and watchlist matching creates vulnerability for asylum seekers with prior asylum attempts or deportations.
Government platform authorization indicates identity data flows to immigration enforcement agencies.
Content exposes surveillance infrastructure that enables government to screen participants in political and civic participation. Watchlist matching and intelligence integration create risk that political participation will be monitored and used to inform law enforcement decisions.
FW Ratio: 50%
Observable Facts
Document reveals system tags users with active intelligence program codenames.
Integration with ICE and federal agencies suggests potential law enforcement use of participation data.
Continuous re-screening every few weeks suggests ongoing monitoring for changes that might trigger law enforcement interest.
Inferences
Intelligence tagging creates risk of surveillance based on protected political expression.
Government agency integration suggests political participation data may be shared with law enforcement.
Automated re-screening indicates systematic monitoring of users for changes in threat status related to political activity.
Content addresses duties and limitations. Authors acknowledge their responsibilities: methodology disclosed as passive recon using only public sources, legal framework cited to justify publication, explicit denial of unlawful access. Authors frame investigation as fulfilling duty to expose rights violations in public interest.
FW Ratio: 50%
Observable Facts
Legal notice states: 'no laws were broken. all findings come from passive recon using public sources - Shodan, CT logs, DNS, HTTP headers, and unauthenticated files served by the target's own web server.'
Authors cite Van Buren v. United States (593 U.S. 374, 2021) and hiQ Labs v. LinkedIn (9th Cir. 2022) to establish legal basis.
Document includes explicit statement: 'the authors are not affiliated with any government, intelligence service, or competitor of any entity named herein. no financial interest. no compensation.'
ADDENDUM acknowledges direct correspondence with Persona CEO and commitment to publishing full written responses.
Inferences
Detailed methodology disclosure demonstrates author commitment to responsible investigation.
Legal citations and limitations show authors respect rule of law while asserting rights.
Transparency about non-affiliation and no financial interest supports good-faith public interest claim.
Engagement with Persona CEO suggests willingness to allow direct response before publication.
Content illustrates discrimination inherent in surveillance system: facial recognition algorithms, watchlist matching, and continuous re-screening create systematic disadvantages based on algorithmic classification rather than individual conduct.
FW Ratio: 50%
Observable Facts
Document describes facial recognition comparing selfies to watchlist photos and tagging users with intelligence codenames.
System continuously re-screens users every few weeks regardless of user conduct or consent.
Source code exposes 14 categories of adverse media screening applied uniformly across users.
Inferences
Continuous automated re-screening without individual justification suggests systematic discrimination through algorithm rather than individual assessment.
Facial recognition scoring creates risk of discrimination based on appearance or algorithmic error.
Universal application of watchlist screening to all users, including those with no adverse history, demonstrates discriminatory system design.
Content demonstrates unequal protection: surveillance system applies facial recognition, watchlist screening, and intelligence tagging without due process, legal justification, or individual assessment. Shows how system creates unequal legal protection for users.
FW Ratio: 50%
Observable Facts
Surveillance system automatically screens users against watchlists using facial recognition algorithms without individual warrants or due process.
Users are continuously re-screened every few weeks without their knowledge or consent.
System applies intelligence codename tagging based on algorithmic matching, not judicial assessment.
Inferences
Automated screening without individual legal justification violates principle of equal legal protection.
Continuous re-screening without due process demonstrates unequal treatment protected by law.
Intelligence codename tagging suggests integration with law enforcement systems without traditional legal safeguards.
Content addresses fair trial rights indirectly by exposing surveillance data that feeds into law enforcement systems. Users are screened against watchlists and tagged with intelligence codenames; this intelligence likely informs law enforcement investigations and potential prosecutions without users' knowledge or ability to challenge algorithmic assessments.
FW Ratio: 50%
Observable Facts
Document describes intelligence codename tagging of users in source code.
System integrates with government platforms (withpersona-gov.com) serving federal agencies.
Integration with Sentry error tracking and analytics on government platform raises questions about data flows to third parties.
Inferences
Intelligence tagging suggests data may be used in law enforcement investigations without users' knowledge.
Government platform integration indicates user data feeds into federal agency decision-making.
Lack of transparency about how watchlist flags inform prosecutorial decisions undermines fair trial principle.
Site structure maximizes distribution of protected speech: open publication, third-party archives, multi-jurisdictional distribution, and dead-drop security arrangements. Authors reserve right to reveal retaliatory actions as additional expression. Structure explicitly protects publication from suppression.
Open publication of privacy violation evidence enables public to understand and contest surveillance. Authors demonstrate commitment to privacy right by publishing detailed infrastructure analysis forcing transparency.
Publication of infrastructure details and integration points allows public to understand and potentially challenge arbitrary surveillance and detention risk.
Publication enables users to understand what biometric property is being collected and used, supporting principle that individuals retain rights to control their data.
Publication enables individuals to understand surveillance risk, supporting principle that freedom of association requires knowledge of who is monitoring.
Open publication and cross-jurisdictional distribution support principle that human rights protection requires international cooperation and shared information.
Publication structure with dead-man's-switch and multi-jurisdictional archives demonstrates commitment to preventing state or corporate suppression of expression.
Site structure supports accessibility of investigative findings through open publication, cross-jurisdictional distribution, and third-party archiving to prevent censorship and preserve access to evidence.
Authors employ dead drops, third-party archives, and pre-positioned key holders to protect physical and informational security. Open publication strategy meant to deter physical threats through transparency.
Publication exposes how identity verification system can be weaponized against asylum seekers and refugees, supporting principle that individuals seeking protection retain rights to privacy and due process.
Publication documents government surveillance infrastructure, supporting principle that political participation must be able to occur without comprehensive monitoring.
Publication includes legal disclaimers, methodology documentation, and author attribution, demonstrating commitment to responsible exercise of expression right.
Publication by independent researchers without government authority demonstrates commitment to legal principle that protection must be equal and applied through transparent, accountable processes.
Publication enables public scrutiny of law enforcement surveillance infrastructure, supporting principle that judicial process must be transparent and subject to public oversight.
Repeated use of emotional framing: 'the watchers,' 'identity surveillance machine,' 'door wide open,' 'rabbit hole deep dive,' 'the messenger does not kill the message.'
appeal to fear
Security notice emphasizing author safety: 'all authors of this document are in good health, of sound mind, and have no plans to hurt themselves, disappear, or die unexpectedly' with dead-man's-switch framing.
causal oversimplification
Framing surveillance system as monolithic threat: 'a facial recognition algorithm is checking whether you look like a politically exposed person' conflates technical possibility with inevitable practice.
Evaluated by llama-3.3-70b-wai: +0.70 (Strong positive)
reasoning
Exposing surveillance abuses
2026-02-26 23:06
eval_success
Light evaluated: Strong negative (-0.80)
--
2026-02-26 23:06
eval
Evaluated by llama-4-scout-wai: -0.80 (Strong negative)
2026-02-26 20:12
dlq
Dead-lettered after 1 attempts: OpenAI, the US government and Persona built an identity surveillance machine
--
2026-02-26 20:10
rate_limit
OpenRouter rate limited (429) model=llama-3.3-70b
--
2026-02-26 20:09
rate_limit
OpenRouter rate limited (429) model=llama-3.3-70b
--
2026-02-26 20:07
rate_limit
OpenRouter rate limited (429) model=llama-3.3-70b
--
2026-02-26 17:41
dlq
Dead-lettered after 1 attempts: OpenAI, the US government and Persona built an identity surveillance machine
--
2026-02-26 17:39
rate_limit
OpenRouter rate limited (429) model=llama-3.3-70b
--
2026-02-26 17:38
rate_limit
OpenRouter rate limited (429) model=llama-3.3-70b
--
2026-02-26 17:37
rate_limit
OpenRouter rate limited (429) model=llama-3.3-70b
--
2026-02-26 17:35
dlq
Dead-lettered after 1 attempts: OpenAI, the US government and Persona built an identity surveillance machine
--
2026-02-26 17:33
rate_limit
OpenRouter rate limited (429) model=llama-3.3-70b
--
2026-02-26 17:33
eval_retry
OpenRouter API error 402 model=llama-3.3-70b
--
2026-02-26 17:33
eval_failure
Evaluation failed: Error: OpenRouter API error 402: {"error":{"message":"Provider returned error","code":402,"metadata":{"raw":"{\"error\":\"API key USD spend limit exceeded. Your account may still have USD balance, but
--
2026-02-26 17:32
rate_limit
OpenRouter rate limited (429) model=llama-3.3-70b
--
2026-02-26 09:02
eval_success
Evaluated: Neutral (0.04)
--
2026-02-26 09:02
eval
Evaluated by deepseek-v3.2: +0.04 (Neutral) 14,918 tokens
2026-02-26 09:00
dlq
Dead-lettered after 1 attempts: OpenAI, the US government and Persona built an identity surveillance machine
--
2026-02-26 09:00
dlq
Dead-lettered after 1 attempts: OpenAI, the US government and Persona built an identity surveillance machine
--
2026-02-26 08:59
dlq
Dead-lettered after 1 attempts: OpenAI, the US government and Persona built an identity surveillance machine
--
2026-02-26 04:23
eval
Evaluated by claude-haiku-4-5-20251001: +0.60 (Moderate positive) 17,677 tokens-0.16
2026-02-26 04:11
eval
Evaluated by claude-haiku-4-5-20251001: +0.75 (Strong positive) 18,826 tokens-0.03
2026-02-26 03:17
eval
Evaluated by claude-haiku-4-5-20251001: +0.78 (Strong positive) 18,045 tokens