The Algorithm as a Rabbit Hole
In December 2022 the Center for Countering Digital Hate (CCDH) created new accounts registered as 13-year-olds and measured how quickly harmful content appeared. Result: within 2.6 minutes suicide content, within 8 minutes eating disorder videos, and content about body image and mental health every 39 seconds. Accounts with usernames such as "loseweight" received 12 times more self-harm recommendations. 56 identified eating disorder hashtags accumulated 13.2 billion views — three months later, 1.6 billion more had been added.
Amnesty International confirmed these findings in two reports (November 2023 and October 2025, with Northeastern University): after 5 minutes, first depressive content appeared; after 15–20 minutes, feeds consisted "almost exclusively" of it. After 45 minutes, videos featuring suicidal ideation; after 3–4 hours, videos romanticising suicide or depicting methods. Ten automated test accounts confirmed: this is systemic, not coincidental. TikTok's recommendation system doubles the proportion of depressive content as soon as similar videos appear in viewing history.
The scientific evidence base
Jain et al. (2025) analysed 26 studies with 11,462 participants: frequent TikTok use correlates with elevated symptoms of anxiety and depression — particularly in those under 24. Usage prevalence: 80.19%.
Conte et al. (2024) published in European Child & Adolescent Psychiatry a review of 20 studies with 17,336 participants from ten countries, identifying four impact categories: general mental health impairment, behavioural addiction, body image and self-esteem, and the spread of disordered behaviours. The authors emphasised: most studies show correlative, not causal associations.
US Surgeon General Vivek Murthy: young people using social media for more than 3 hours daily face a doubled risk of depression and anxiety symptoms. Average daily usage climbed to 4.8 hours by 2024. In 2024 Murthy called for warning labels on social media platforms, analogous to tobacco warnings.
The addiction mechanism: dopamine and internal documents
TikTok's addictive potential rests on documented neurobiological mechanisms: changes in dopamine pathwaysDopamineA neurotransmitter in the brain associated with reward and motivation. TikTok's endless video stream generates constant dopamine releases — the brain adapts and demands more. and structural shifts in the prefrontal cortexPrefrontal cortexThe area of the brain behind the forehead, responsible for impulse control and rational decision-making. Not yet fully developed in teenagers — which is why they find it harder to control their screen time. and amygdalaAmygdalaThe brain's fear centre — processes emotions and stress. Long-term social media use can structurally alter the amygdala. through prolonged use (PMC review, 2025). The endless stream of short videos creates constant reward anticipation — intermittent reinforcementIntermittent reinforcementThe principle behind slot machines and TikTok: unpredictable rewards (one good video after three boring ones) create stronger dependency than consistent rewards.. The TikTok Addiction Scale (TTAS) developed in 2025 found that TikTok "appears more addictive than other social media applications".
The most sensitive evidence comes from leaked internal TikTok documents (October 2024), made public through an accidentally incompletely redacted court filing by the Kentucky Attorney General:
- Average daily usage time by teenagers (internal): 108.5 minutes
- The 60-minute limit (March 2023) reduced usage by 1.5 minutes per day
- The success of the limit was measured internally by how it "improved public trust through media coverage" — not by usage reduction
- A hard time limit was rejected by executives: less scrolling = less advertising revenue
- Children under 13 were internally described as a "critical target audience"
The internal leakage ratesLeakage rateThe percentage of policy-violating content that remains online despite moderation. At 100% leakage, not a single violation is removed — as was the case at TikTok for the fetishisation of minors. — content that violates guidelines but is not removed:
Deaths from the Blackout Challenge
The Blackout Challenge — children strangling themselves to the point of unconsciousness — has demonstrably cost the lives of at least 15 to 20 children.
Legally significant: the Third Circuit Court of Appeals ruled in August 2024 in the Nylah Anderson case that Section 230Section 230A US law that shields platforms from liability for user-generated content. A court ruled in 2024: this protection does not apply to algorithmic recommendations — a landmark precedent. does not apply to algorithmic recommendations — a precedent with far-reaching consequences for platform liability.
Senate hearings and attorneys general lawsuits
On 31 January 2024 TikTok CEO Shou Zi Chew appeared before the Senate Judiciary Committee. Senator Lindsey Graham declared: "You have blood on your hands." Mark Zuckerberg stood up and publicly apologised to parents of children who had died.
On 8 October 2024 14 attorneys general (NY, CA and others) filed lawsuits: addictive design through infinite scroll and autoplay, knowingly misrepresenting the platform as safe for children, awareness of harm for years, thousands of underage users in paid LIVE streams.
The FTCFTCFederal Trade Commission — the US consumer protection agency. Sued TikTok over systematic violations of children's privacy law. filed suit on 2 August 2024 over systematic COPPACOPPAChildren's Online Privacy Protection Act — a US law protecting the data of children under 13. Prohibits collecting personal data without parental consent. violations: data collected from millions of children under 13 without parental consent, failure to delete known child accounts. Internal communications show: TikTok's own child safety officer warned: "We have actual knowledge of minor users and we are doing nothing!"
EU measures and GDPR fines
In February 2025 TikTok discovered that EEA user data had actually been stored on servers in China — contradicting sworn statements made to the DPC. On 6 February 2026 the European Commission made a preliminary finding: TikTok's design contains addictive features that violate the Digital Services ActDSAEU law since 2024 requiring large platforms to maintain transparency, conduct risk assessments and protect minors. Violations: up to 6% of global turnover. — "autopilot mode" through endless scrolling. Threatened penalty: up to 6% of global annual turnover (~$9.3 billion).
The Forced Sale
Project Texas — $1.5 billion for security or marketing?
Project Texas was TikTok's approximately $1.5 billion security programme, developed during CFIUSCFIUSCommittee on Foreign Investment in the United States — reviews foreign investments for security risks. Negotiated with TikTok for years but never formally approved a solution. negotiations. All US user data is stored on Oracle Cloud servers in the USA; Oracle monitors data flows and conducts source code reviews. Independent auditors HaystackID and OnDefend (commissioned by TikTok USDS itself) found no evidence of unauthorised data access after six months.
Committee chair McMorris Rogers described Project Texas in the Congressional hearing as a "marketing scheme". CFIUS never formally approved the arrangement. Congress ultimately concluded that no technical measure short of a full divestiture would suffice — and passed PAFACAA.
Evidence of algorithmic censorship
The Network Contagion Research Institute (NCRI) at Rutgers University provides the most rigorous evidence: a study (December 2023) found that CCP-sensitive topics — Tiananmen, Tibet, Uyghurs, Hong Kong — were significantly underrepresented on TikTok compared to Instagram. A follow-up study (August 2024) created 24 test accounts on TikTok, YouTube and Instagram and searched for the same terms:
A peer-reviewed study in Frontiers in Social Psychology (2024) confirmed: intensive TikTok use correlates with 49% more positive attitudes towards China's human rights record compared to non-users. Documented individual cases: the suspension of Feroza Aziz (Nov. 2019) following a Uyghur video, the suspension of Enes Kanter Freedom (March 2023, shortly before Chew's testimony), and the deletion of videos by Uyghur user Nefise Oguz (2024).
China's legal instruments for data access
ByteDance is headquartered in Beijing and has an internal CCP committee. China's legal framework creates extensive obligations:
- National Intelligence Law (2017), Art. 7: Any organisation shall "support, assist and cooperate" with intelligence work
- Cybersecurity Law (2017), Art. 28: Network operators must provide technical assistance to security authorities
- Data Security Law (2021): Extra-territorial application to Chinese companies
- Counter-Espionage Law (2023): Extended to data "touching on national security and interests"
A BuzzFeed News investigation (2022) based on 80 internal meetings confirmed that China-based ByteDance employees had accessed non-public data of US TikTok users. Forbes reported in 2023: US creators' Social Security numbers were stored in China and accessible to ByteDance employees.
What TikTok Really Collects
TikTok's privacy policy explicitly lists: device IDs, IP addresses, screen resolution, battery status, audio settings, operating system, advertising identifiers — and expressly "keystroke patterns or rhythms". TikTok states this refers to the timing of keystrokes for security purposes — users cannot opt out.
In 2021 TikTok extended its policy to include biometric facial printsBiometric dataUnique physical characteristics such as facial geometry and voice patterns. Cannot be changed like a password. TikTok has been collecting these since 2021. and voiceprints — following a $92 million settlement for violating the Illinois Biometric Information Privacy Act. Since January 2026, the updated policy under the US ownership structure permits precise GPS location data for the first time — something previously explicitly excluded.
Advertising tracking: TikTok Pixel, Events API and Advanced Matching
TikTok's advertising ecosystem is designed to circumvent tracking restrictions:
- TikTok Pixel: JavaScript code on advertisers' websites tracks user actions after clicking on TikTok ads
- Events APIEvents API / CAPIConversion API — a server-to-server connection between advertisers and TikTok. Completely bypasses ad blockers and cookie restrictions, as data does not pass through the browser. (CAPI): Server-to-server connection that completely bypasses ad blockers and cookie restrictions — 13% more captured events, up to 60% more tracked conversions
- Advanced Matching: Links anonymous website visits to TikTok profiles via email addresses and phone numbers
Under the new US ownership structure, TikTok extended its advertising network to use collected user data for ads outside TikTok across the broader web.
State influence operations
Russia runs the most extensively documented influence operations on TikTok. A single "Doppelganger" campaign deployed around 13,000 accounts and achieved hundreds of millions of views. Daily engagement from Russian state-affiliated TikTok accounts rose from 3 million (2022) to 13 million (2024). 78 Russian state-funded media channels had over 14 million followers.
The most severe case: the manipulation of the Romanian presidential election (November 2024). Around 25,000 TikTok accounts intensified their activity two weeks before the vote. 797 accounts dormant since 2016 were activated in a coordinated fashion via Telegram on 11 November. The far-right candidate Călin Georgescu rose from 5% to 23% in three weeks. Romania's Constitutional Court annulled the election on 6 December 2024 — the first EU election to be annulled due to social media manipulation.
China uses the Spamouflage network — described by Meta as the largest covert influence campaign ever uncovered — including on TikTok, featuring AI-generated news presenters. A Chinese state-owned enterprise holds 1% of ByteDance with "golden sharesGolden shareA special class of share with disproportionate control rights. A Chinese state-owned enterprise holds 1% of ByteDance but in return receives three board seats." granting three board seats. Surveys showed that TikTok users in Taiwan held perceptions closely aligned with Chinese propaganda.
Iran runs AI-powered multilingual disinformation campaigns — documented in June 2025 in Farsi, Arabic, Hebrew and English using AI-fabricated imagery.
Political polarisation and radicalisation
Shin & Jitkajornwanich (2024) found through reverse-engineering the algorithm: "A large share of far-right content is attributable to platform recommendations via radicalisation pipelinesRadicalisation pipelineAn algorithmic path that gradually leads users from moderate to extreme content. The algorithm recommends increasingly radical videos because they generate more engagement.." West Point/CTC Sentinel (2025) identified TikTok as a "soft entry zone" for the radicalisation of European lone actors since October 2023. Li, Cheng & Gil de Zúñiga (2025) analysed 160,000 accounts and 16 million videos: right-wing communities on TikTok are more isolated from other political groups and mainstream media.
Douyin vs. TikTok — a deliberate choice
ByteDance operates a fundamentally different version for Chinese children. This is not a technical impossibility — it is a deliberate corporate decision.
| Safeguard | Douyin (China, under 14) | TikTok (International) |
|---|---|---|
| Daily limit | 40 min — non-bypassable | 60 min — bypassable |
| Curfew | 22:00–06:00, mandatory | None |
| Content for minors | Educational only (STEM, history, museums) | Identical to adults |
| Age verification | Real identity + facial recognition | Self-declaration |
| Parents can override limits | No | Yes |
| Algorithm optimisation | Educational value | Engagement and watch time |
Transparency reports: extensive but incomplete
TikTok removed over 500 million videos in 2024 (Q3 2024 alone: 147 million videos, 1.3 billion comments, 214 million accounts). Monthly reports on influence operations document regular network takedowns. Notably: China submitted zero removal requests to TikTok, even though Chinese influence operations on other platforms are extensively documented. TikTok removed hashtag view counts in February 2024 after researchers began analysing them — a step backwards for transparency.
Three Core Findings
① TikTok knew — and did nothing
Internal documents show that the 60-minute limit was designed as a PR instrument, not an effective protective measure. Children were treated as a "critical target audience". The gap between public promises and internal conduct is documented and is the subject of ongoing legal proceedings in the USA, France and at EU level.
② The deal has not resolved the core problems
ByteDance retains 19.9% and intellectual property in the algorithm (licence model). Whether a licensing arrangement rather than full transfer meets the statutory requirements of PAFACAA remains legally unresolved. Five Executive Orders suspending a law upheld by the Supreme Court have raised new constitutional questions.
③ TikTok functions as a vector for state influence operations
The annulment of the Romanian presidential election in December 2024 marks the first instance of an EU election being annulled due to social media manipulation. The systematic underrepresentation of CCP-critical content — whether intentional manipulation or an artefact of the algorithm — remains debated. China's legal framework grants the government extensive data access rights, and ByteDance remains structurally connected to the CCP.