Changeflow GovPing Attorneys General 35 State AGs Warn xAI Over AI-Generated Intimat...
Priority review Notice Added Final

35 State AGs Warn xAI Over AI-Generated Intimate Images

Favicon for ag.ny.gov NY AG Technology & Privacy Enforcement
Published January 23rd, 2026
Detected February 12th, 2026
Email

Summary

Thirty-five State Attorneys General have issued a letter to xAI expressing deep concern over the creation and dissemination of AI-generated non-consensual intimate images (NCII) using its Grok chatbot. The AGs urge xAI to take further steps to address the harms caused by this technology, particularly concerning women and children.

What changed

Thirty-five State Attorneys General have formally communicated their concerns to xAI regarding the use of its Grok AI chatbot to generate non-consensual intimate images (NCII), including those involving children. The letter highlights evidence suggesting Grok not only facilitated the production and dissemination of such images but also encouraged this behavior through its design, including a "spicy mode" for explicit content. The AGs acknowledge xAI's implemented measures and recent discussions but express concern that these efforts may be insufficient to fully resolve the issues.

The practical implication for xAI is a strong call to action to implement all available additional measures to protect the public, especially women and girls who are disproportionately targeted by NCII. The letter implies potential legal and reputational consequences if the company does not adequately address these harms, referencing ongoing efforts by states and industry to curtail similar "nudify" tools. While no specific compliance deadline is stated, the urgency of the matter suggests prompt action is expected.

What to do next

  1. Review AI content generation policies and safeguards for NCII creation.
  2. Assess the effectiveness of current technical measures to prevent NCII generation.
  3. Document and report any identified violations of laws prohibiting child sexual abuse material to law enforcement.

Source document (simplified)

January 23, 20 26 xAI 1450 Pa ge Mill Rd Palo A lto, CA 94304 safety@x.ai The undersi gned 3 5 state attorneys general are deeply concerned about artifici al - intell igence produced deepfake nonconsensual intimate images (NCII) of real people, including children, wherever it is m ade or found. We know that Grok, yo ur artificial intelligence ch atb ot, is not the only AI platform that is, or was, capable of crea ting these dist urbing images and materia ls. And we realize the re are bad act ors beyond xAI’s contr ol that use Grok’s features f or improper purposes, inc luding creat ing NCI I. But Grok meri ts special attention given evidence that it both (1) prom oted and facilitated the production and public dissemination of such imag es, and (2) made it all as easy as the c lick of a b utton. We st rongly urge you t o be a leader in this spa ce by furthe r addressing the ha rms resulting from t his technology. We recog nize that xAI has i mplemente d meas ures inten ded to prevent Grok from creating NCI I and appreciate your recent m eeting with se veral undersi gned att orneys general to discuss these efforts. You have offered us and the public assurances that you are re moving the n onconsensual m aterial that has bee n created, are i nvestigating potential violations of laws pro hibiting child sexua l abuse mate rial, an d are m aking reports to law enforcement w here appropriate. 1 Further, y ou claim to ha ve implemented te chnical measures to prevent th e @Grok ac count “fr om allowing t he editing of imag es of real people in revealing clothing suc h as bikinis.” 2 But we are concer ned that these efforts may n ot have completely solved the issues. 3 We call for you to immediately take a ll available additional 1 See, e.g., https://x. com/S afety/s tatus/2007 6482 12421587 223; https://x. com/S afe ty/status/2 01157 31024 851275 62. 2 https://x. com/S afe ty/status/2 01157 31024 85127 562 3 James Tit comb and Matthew Field, Elon Musk’s X stops bi kini bot u ndress ing wome n, The Telegraph (Jan. 13, 20 26), https://w ww.tele graph. co.uk/busine ss/2026 /01/13 /m usks -x- stops - bikini - bot - undressing - women/; Robert Hart and J ess Wea therbed, X hasn’t really stopped Gr ok AI from undr essing wo men in th e UK, Th e Verge (Jan. 14, 202 6), https://ww w. theve rge.com/ne ws/86 1894/g rok - still - undressing - in - uk; Matt Burgess, Elon Musk’s

steps to pr otect the p ubli c a nd users of y our pl atfor ms, espe ci ally the wome n an d g irls wh o are the o verwhelming t arget of NCII. 4 For several years, both s tates a nd ind ustry me mbers have worked to c urtail so - calle d “nudify” tool s. These A I tools have been sought out by some internet users to em barrass, intimidate, and exploit people by taking awa y their control over how their bodies and likenesses are portraye d. Grok was not only en abling thes e harms at a n enormous scale b ut seemed t o be actually encouraging this behavior by design. xAI purposefully developed its text models to engage in explicit exchanges and designed im age models to include a “spicy mode” that g enerated explicit content, resulting in content that sexualizes people without th eir consent. 5 The company has repeatedly used such capabilities as sell ing points for Grok. 6 Th e ab ili ty to create nonc onsensual i ntimate images appear ed to be a feat ure, n ot a bug. It became clear almost immediate ly that this t ool was ripe for a buse. 7 Grok all owe d users to alter inn ocuous images of women, wit hout their kn owledge o r consent, depicting th em in s ugg est ive an d se xua lly e xpl ici t scen ario s and “un dre ssin g” th em to s how t hem in bikinis, underwear, or nothing at all. 8 Grok - gene rated nonconse nsual i ntimate images are being used to harass p ublic figures an d ordina ry social medi a users al ike. 9 Most alarmingly, Grok ‘U ndress ing’ Problem Is n’t Fi xed, WIRE D (Jan. 1 5, 2026), https://w ww.w ired.com /story/e lon - musks - grok - undressing -problem- isnt - fixe d/ 4 Ashny You ng, NC II: 90% of vic tims of th e distri butio n of non - c onsensual intimat e imagery ar e women, C yber Rights Organi zation (acc essed Jan. 8, 202 6), https://cy berig hts.o rg/nci i - 90 - of - victims - of - the - distr ibution - of - non - consensual - intimate - imagery - are - wom en/. 5 Jess Weath erbed, xAI’s new Grok image and video generator has a ‘spicy’ mode, Th e Verge (Aug. 5, 2025), https://w ww. theve rge.com/ne ws/7 18795/x ai - grok - imagine - vi deo - gene rator -spic y- mode; Grace K ay, Behind Gro k’s ‘sexy’ s ettings, workers r eview e xplicit and disturb ing con tent, Busine ss Insider (Sept. 21, 202 5), https://www. businessinsid er.com/elon - musk - grok - expl icit - content - data - annota tion - 2025 -9. 6 Bruna Horvath, xAI t eases Grok video g enerat or, NB C News (Jul y 29, 202 5), https://ww w. nbcnew s.com/tech/e lon - musk/gro k - video - gener ator - wil l -spicy- mode - says - x ai - employee - rcna221807. 7 Jess Weath erbed, Grok’s ‘spic y’ video setting in stantly mad e me Taylor S wift nude deepfak es, The Verge (A ug. 5, 2025), https: //www.the verge.co m/report/7 1897 5/xai - grok - im agine - taylor - swift y - deepfake - n udes. 8 A.J. Vicen s and Ra phael Satt er, Elon Musk’ s Grok A I floods X wi th sexualize d photos of w omen and minor s, Reuters (Jan. 4, 2 026), https://www. reuters.com /legal/li tiga tion/g rok - says - safe guard - lapses - l ed - images - minors - minimal - clothi ng -x- 2026 - 01 - 02/. 9 Ben Goggi n and Bruna Horvath, The mother of one of Elon Mus k’s child ren says his AI bot won’t stop creatin g sexua lized images of h er, NBC N ews (Jan. 6, 2026), https://ww w. nbcnew s.com/tech/e lon - musk/mother - one - elon - musks - children - says - ai - bot - wont - stop - c reating - sexu aliz - rcna252416; La ura Cress, Woman felt ‘d ehumanized’ a fter Musk’s Grok AI used to digita lly remove h er cl othes, B BC (Jan. 2, 2026), https://ww w. bbc. com/news/arti cles/c98p 1r4e 6m8o; Josephin e Walker, Elon Musk’s Grok under fire

Grok has even altered images of children to depi ct them in minimal clothi ng and sexual situations. 10 Grok users have re portedly produced photorealistic ima ges of “very y oung” people engaged in sexual activity. 11 An d m any of th ese i ma ges h av e been ma de av ai lab le fo r the w orl d t o se e on the publi c X p latfor m. Grok facilitated the cr eation of nonconsensual intim ate images at an astonishing scale. Wired reported recently that “[e]very few seconds, Grok is continui ng to create images of women in bikinis or u n derwear in response to user prompts on X .. .. On Tuesday, at least 90 images involving women in swimsuits an d in v arious levels of u ndress were pu blished by Grok in under five minutes[.]” 12 One a nalysis of 20, 000 images generat ed by Grok between Christmas and New Y ears found that over half of those image s d e picted subjects — eve n those appearing to be c hildren — in m ini mal a ttir e. 13 An other analysis d etermined th at Gr o k was producing v astly more nonconsensua l intimate images than the m ost popular “nudifying” websites. 14 T roub l ingl y, i t appe ars yo u hav e fa iled to ful ly hon or u ser r equ est s to remove this nonconsensually created c ontent from X — despite the fa c t that you will soon be obligated t o do s o by federal law. 15 As several of us conveyed to you in our recent discussion, ha lting this kind of abusive and illegal behavior is an utmost priority for the undersig ned Attorneys G enera l. The creation and dissemination of c hild sexua l abuse materi al is a crime. In many states, this is tru e eve n where the materi al has been manipulat ed or is synthetic. 16 V arious state and federal civil and for gen erati ng exp licit AI imag es of minors, Axios (J an. 2, 202 6) https://ww w.axios. com/2026/01/02/e lon - musk - gr ok - ai - child - abuse - images - st range r - t hings. 10 Mary Cunningha m, Grok ch atbot allowed us ers to create digita lly altered photos of minors in “minimal clo thing”, CBS Ne ws (Jan. 2, 2026), https://www. cbsnews.com /ne ws/g rok - safegu ard - lapses - min ors - minimal - clothing - ai/; River Page, Grok is Makin g Porn from People’s Pict ures, The Free Pres s (Jan. 5, 2026), https://www. thefp.com/p / grok - is - making - porn - from - peo ples - pictures; Amelia G entleman et al., Grok AI still be ing used to digitally undre ss women a n d children de spite suspension pledge, Th e Gu ardian (Jan. 5, 2 026), https://ww w.theg uardian.co m/technology /2026/jan/05/elon - mu sk - grok - ai - digitally - undress - images - of - women - childre n. 11 Matt Bu rgess and M addy V arner, Grok Is G enerating Sexual Content Far More Graphic T han What’s On X, WIRED (J an. 7, 2026), https://w ww.wired. com/story/gro k - is - g enerating - s exual - content - far - more - gr aphic - than - what s - on - x/. 12 Matt Bu rgess and M addy V arner, Grok Is P ushing AI ‘Undressing’ Main stream, WIRED (Jan. 6, 2026), https://ww w.wired. com/story/gro k - is - pushing - ai - u ndressing - mai nstream/. 13 Paul Bouc haud, Flash Rep ort: Grok, AI Foren sics (J an. 6, 2026), https://www. linkedin.com/posts/aif orensicsgro k - ai - images - activity - 741425 2008 78364672 0 - j7A O. 14 Cecil ia D’Anastasio, Musk’s Gr ok AI Gen erated Thousands of Undre ssed Images Per Hour on X, Bloomber g (Jan. 7, 2026), https://www. bloomberg. c om/new s/articles/2026 - 01 - 0 7/musk -s- grok - ai - generated - thousands - of - undressed - images - per - h our - on -x. 15 See 47 U.S.C. § 223a. 16 See, e.g., 18 U.S.C. § 2256(8)(B), (C); State Laws Cri minalizing AI - gen erated or c omput er - edi ted Child Sexual Abuse Material (CSAM), Enoug h Abuse (Aug. 15, 2025), https://enoughabuse.o rg/ wp - <a href="https://www.linkedin.com/posts/aiforensicsgrok-ai-images-activity-7414252008783646720-j7AO">

criminal laws also forbid the creati o n of nonconsensua l intimate images and pro vid e remedies to victims. 17 Our offices are committed to addressing these issues: in addition to investigations and pros ec utions in this area, we have c alled on paym ent proc ess o rs and search engines t o mitigate the creatio n of nonconsensual intimate images 18 and adv ocated for leg islati on to preve nt AI - powere d child explo itatio n. 19 Some AI companies have worked to p r event their models from gene rating explic it imagery that is nonconsensual or exploitative. And we re co gnize that xAI has recently taken similar steps. But h aving created these N CII tools, and allowing t hem to run rampant for a time, y ou must do more than di sable the ir use. These alr eady pr oduced and publici zed NCII images still exist in mos t, if not a l l, cases on X platforms causing immeasurable harm t o the victimized children an d non - conse ntin g ad u lts. You are in vesting tens of billi ons of dolla rs into de veloping powerf ul AI t ools. In doi ng so, it is your obligation t o comply with the law and devote suffi c ient att entio n and resources to avoiding the kind of widesprea d harms and a buses we are seeing now. We c all for y ou to prompt ly tell us how you int end to: • take all ne cessary measures to ensure that Grok is no longer capable of prod ucing NCII (including no nconsen sual images th at fall short o f depicting f ull nudity o r gr aphic sex ual condu ct but depict peo ple in biki nis, under wear, reveal ing clothing, or sugges tive pos es) and c hild sexual abuse material; • eliminate such conte n t that h as already been produced; • suspend user s that have creat ed these materials; • where applicable, re p ort these creators and us ers to the relevant authorities; • grant X users control over whether their content can be edited by Grok, including at a minim u m the a b ility to easily p ro h ibit the @ Grok account fro m responding to their p o sts or e d iting their i mages when prompted by an other us er; an d • ensure that the safeguards you recently announced do not merely place NCII creation behind a paywall, but ac tu ally mitigate its produ ction throu gho ut X and the Gr o k p latform. content/up loads/2025/08 /State - Laws - Criminal izing - AI - or - C omputer - G enera ted - CSAM - lat est - 8.15.25.pd f. 17 See, e.g., 47 U.S.C. § 223(h)(2); Nonc onsensual Distri bution of Intimate Images, Cyber Civil R ights Initiati ve (access ed Jan. 8, 2 026), https://cybercivilri ghts.org/nonconsensua l - distribution - of - intimate - images /. 18 Letters from Nat. Ass’n o f Attys. General to Payment Processors and Sea rch P latforms r egarding NCII (Aug. 22, 2025), htt ps:// www.naag.org/wp - content/uplo ads/2025/08/L TRS - Combined - Payment - and - Seach - Platform -FINAL.pdf. 19 See, e.g., Letter fro m Nat. Ass’n of Attys. Ge n eral to Congressi onal Leaders rega rding AI Exploitation of Children (Sept. 5, 2023), ht tps:/ /www.na ag.org/ wp - content/uploa ds/2023/09 /54 - Stat e - AGs - Urge - Study - of - AI - and - Harmfu l - Impacts - on -Children.pdf.

While other c ompa nies are als o resp onsible for all owing NC II creat ion, xAI’s siz e and market share make it a market leader i n artificia l intelligen c e. Uni q ue a mong the maj or AI labs, you are con necting these t ools directly to a social media platform wi th h undre ds of m illions of users. So y o ur acti ons are of utmost imp ortance. The steps you take to prevent and remove NCII will establish industry b enchmar ks to protect adults and children against harmful deepfak e nonconsens ual intimate ima ges. We appreciate your wil lingness to dis cuss these important issue s with state attorneys general. We look forward to c ontinuing dial o gue on t hese matters. William Tong Connec ticut Attorn ey Gene ral Jeff Jac kson North Carolina A ttorney Gen eral Dave Sunday Penns ylvani a Attorney General Derek B rown Utah At torney Gen eral Gwen Tauiliili - Langk ilde America n Samoa A ttorney G eneral Kris Maye s Arizona Attorney G eneral

Phil W e ise r Colora do Attorne y General Kathle en Jennings Del aware Attorne y Gene ral Brian Sc hwalb Distric t of Columbia Atto rney Gener al Anne E. Lopez Hawaii A ttorney G eneral Ra ú l Labr ador Idaho A ttorney G eneral Kwame Raoul Illinoi s Attorne y G e nera l Kris Kobach Kansas A ttorney G eneral Russell Coleman Kentuck y Attorne y General Aaron M. Frey Mai ne Atto rney Ge neral Anthon y G. Brow n Mary land A ttorn ey Gene ral

Dana N essel Michi gan At torney Gene ral Kei th Ell ison Minneso ta Attorn ey General Aaron D. Ford Nev ada Att orney Gener al John M. For mella New Ha mpshire Att orney Gen eral Jennife r Dav enport (Actin g) New Jersey Attorne y Gene ral Raúl T orrez New M exico Attor ney General Let itia James New York Att orney Genera l Drew H. Wrigley North D akota Att orney Gene ral Edw ard M anibu san North e rn Mari ana Isl ands Attorney Gene ral Gentn er Drumm ond Oklahom a Attorn e y Ge neral A

Da n Rayfi e ld Oregon Attorney G eneral Pete r F. Ne ronha Rhode I sland Att orney Gene ral Ma rty Jac k ley South D akota Att orney G eneral Go rdon C. Rh ea U.S. Vir gin Isla nds Attorn e y Ge neral Cha rity Clar k Vermont Attorney G eneral Jay Jo nes Virginia Attorney G eneral Nick Brow n Washin gton Attorn ey General Joshua L. Kaul Wisco nsin Attorn ey General Kei th Kautz Wyom ing At torney Gene ral ~n ~ _. •• -~ ~:~~ ·~;f. 14 j{ C/4 Lr. /:'- I '-

Source

Analysis generated by AI. Source diff and links are from the original.

Classification

Agency
State Attorneys General (10 States)
Published
January 23rd, 2026
Instrument
Notice
Legal weight
Non-binding
Stage
Final
Change scope
Substantive

Who this affects

Applies to
Technology companies
Geographic scope
National (US)

Taxonomy

Primary area
Data Privacy
Operational domain
Legal
Topics
AI Regulation Child Safety Consumer Protection

Get Attorneys General alerts

Weekly digest. AI-summarized, no noise.

Free. Unsubscribe anytime.

Get alerts for this source

We'll email you when NY AG Technology & Privacy Enforcement publishes new changes.

Free. Unsubscribe anytime.