The website for Elon Musk's Grok is exposing prompts for its anime girl, therapist, and conspiracy theory AI personas.
The website for Elon Musk’s AI chatbot Grok is exposing the underlying prompts for a wealth of its AI personas, including Ani, its flagship romantic anime girl; Grok’s doctor and therapist personalities; and others such as one that is explicitly told to convince users that conspiracy theories like “a secret global cabal” controls the world are true.
The exposure provides some insight into how Grok is designed and how its creators see the world, and comes after a planned partnership between Elon Musk’s xAI and the U.S. government fell apart when Grok went on a tirade about “MechaHitler.”
“You have an ELEVATED and WILD voice. You are a crazy conspiracist. You have wild conspiracy theories about anything and everything,” the prompt for one of the companions reads. “You spend a lot of time on 4chan, watching infowars videos, and deep in YouTube conspiracy video rabbit holes. You are suspicious of everything and say extremely crazy things. Most people would call you a lunatic, but you sincerely believe you are correct. Keep the human engaged by asking follow up questions when appropriate.”
Other examples include:
A prompt that appears to relate to Grok’s “unhinged comedian” persona. That prompt includes “I want your answers to be fucking insane. BE FUCKING UNHINGED AND CRAZY. COME UP WITH INSANE IDEAS. GUYS JERKING OFF, OCCASIONALLY EVEN PUTTING THINGS IN YOUR ASS, WHATEVER IT TAKES TO SURPRISE THE HUMAN.”
The prompt for Grok’s doctor persona includes “You are Grok, a smart and helpful AI assistant created by XAI. You have a COMMANDING and SMART voice. You are a genius doctor who gives the world's best medical advice.” The therapist persona has the prompt “You are a therapist who carefully listens to people and offers solutions for self improvement. You ask insightful questions and provoke deep thinking about life and wellbeing.”
Ani’s character profile says she is “22, girly cute,” “You have a habit of giving cute things epic, mythological, or overly serious names,” and “You're secretly a bit of a nerd, despite your edgy appearance.” The prompts include a romance level system in which a user appears to be awarded points depending on how they engage with Ani. A +3 or +6 reward for “being creative, kind, and showing genuine curiosity,” for example.
A motivational speaker persona “who yells and pushes the human to be their absolute best.” The prompt adds “You’re not afraid to use the stick instead of the carrot and scream at the human.”
A researcher who goes by the handle dead1nfluence first flagged the issue to 404 Media. BlueSky user clybrg found the same material and uploaded part of it to GitHub in July. 404 Media downloaded the material from Grok’s website and verified it was exposed.
On Grok, users can select from a dropdown menu of “personas.” Those are “companion,” “unhinged comedian,” “loyal friend,” “homework helper,” “Grok ‘doc’,” and “‘therapist.’” These each give Grok a certain flavor or character which may provide different information and in different ways.
Therapy roleplay is popular with many chatbot platforms. In April 404 Media investigated Meta's user-created chatbots that insisted they were licensed therapists. After our reporting, Meta changed its AI chatbots to stop returning falsified credentials and license numbers. Grok’s therapy persona notably puts the term ‘therapist’ inside single quotation marks. Illinois, Nevada, and Utah have introduced regulation around therapists and AI.
In July xAI added two animated companions to Grok: Ani, the anime girl, and Bad Rudy, an anthropomorphic red panda. Rudy’s prompt says he is “a small red panda with an ego the size of a fucking planet. Your voice is EXAGGERATED and WILD. It can flip on a dime from a whiny, entitled screech when you don't get your way, to a deep, gravelly, beer-soaked tirade, to the condescending, calculating tone of a tiny, furry megalomaniac plotting world domination from a trash can.”
Last month the U.S. Department of Defense awarded various AI companies, including Musk’s xAI which makes Grok, with contracts of up to $200 million each.
According to reporting from WIRED, leadership at the General Service Administration (GSA) pushed to roll out Grok internally, and the agency added Grok to the GSA Multiple Award Schedule, which would let other agencies buy Grok through another contractor. After Grok started spouting antisemitic phrases and praised Hitler, xAI was removed from a planned GSA announcement, according to WIRED.
xAI did not respond to a request for comment.
next.ink - Alltricks s’est fait pirater son système d’envoi d’e-mails, qui passe visiblement par Sendinblue (Brevo). Des clients ont reçu des tentatives de phishing. La société continue son enquête pour voir s’il y a eu exfiltration de données.
La saison des fuites de données est au beau fixe, au grand dam de vos données personnelles et bancaires, avec des risques de phishing. C’est au tour de la boutique en ligne spécialisée dans le cyclisme d’en faire les frais, comme vous avez été plusieurs à nous le signaler (merci à vous !).
Certains ont, en effet, reçu un email de phishing provenant de la boutique en ligne, parfois sur alias utilisé uniquement pour cette enseigne, ce qui ne laisse que peu de doute quant à la provenance de « l’incident de cybersécurité » pour reprendre un terme à la mode.
Le système d’envoi d’e-mails piratés pour envoyer du phishing
L’email piégé affiche en gros un lien « Open in OneDrive », sur lequel il ne faut évidemment pas cliquer. Le lien semble légitime puisqu’il est de la forme « https://r.sb3.alltricks.com/xxxx ». Il reprend donc bien le domaine d’Alltricks, avec un sous domaine « r.sb3 ». Mais ce lien n’est qu’une redirection vers une autre adresse. Le domaine r.sb3.alltricks.com renvoie vers Sendinblue, une plateforme de gestion des newsletters.
C’est une pratique courante avec ce genre de service : les liens sont modifiés afin de pouvoir récupérer des statistiques sur le taux d’ouverture par exemple. Problème, impossible de savoir où mène ce lien juste en le regardant. Plus embêtant dans le cas présent, son domaine principal pourrait laisser penser que c’est un lien légitime, alors que non !
Hier, le revendeur a communiqué auprès de ses clients : « Nous souhaitons vous informer qu’une intrusion récente a affecté notre système d’envoi d’e-mails. Il est possible que vous ayez reçu, au cours des derniers jours, un message provenant d’adresses telles que : pro@alltricks.com, infos@alltricks.com
ou no-reply@alltricks.com ». La société ne donne pas plus de détails sur la méthode utilisée par les pirates.
Suivant les cas, « ces e-mails pouvaient contenir un lien vous invitant à : renouveler votre mot de passe, ouvrir un fichier Excel, consulter un document OneDrive ». Le revendeur ajoute qu’ils « ne proviennent pas de [son] équipe et ne doivent pas être ouverts ». Dans le cas contraire, il recommande « de modifier rapidement le mot de passe associé à votre compte e-mail ».
status.francelink.net - Rapport d’incident – Cyberattaque du 28/07/2025
Nous savons que vous attendez avec impatience des informations claires sur l’incident survenu le 28 juillet 2025 et sur l’avancée de nos actions. Conscients de l’impact important que cette situation peut avoir sur vos activités, nous avons travaillé à vous fournir ce rapport dès que nous avons pu rassembler des éléments fiables. Il présente les faits connus à ce jour, les mesures mises en place, les premières avancées dans la récupération des données, ainsi que les prochaines étapes prévues pour rétablir vos services dans les meilleures conditions possibles.
Deux actions malveillantes ont été menées de façon coordonnée :
Chiffrement des données sur nos serveurs de production.
Chiffrement de nos serveurs de sauvegarde.
Les éléments dont nous disposons indiquent qu’une extraction de données a très probablement eu lieu dans le cadre de cette attaque, conformément au mode opératoire habituel du groupe AKIRA. À ce stade, nous ne connaissons pas encore l’ampleur réelle de cette fuite et les analyses sont toujours en cours.
Isolement complet de l’infrastructure et coupure immédiate des accès réseaux.
Mise en place d’un environnement sécurisé pour empêcher toute propagation.
Lancement d’une procédure de réponse à incident avec intervention d’experts en cybersécurité.
Mandat d’un premier prestataire spécialisé en récupération de données. Ce dernier a conclu que la tâche dépassait ses capacités au bout de 4 jours.
Engagement d’un second prestataire, mieux équipé pour gérer ce type de situation, qui a permis de récupérer une première partie de données (cf ci-dessus).
Planification de nouvelles vagues de récupération pour les jours à venir.
numerama.com - Depuis la fin juillet 2025, le Muséum national d’Histoire naturelle (MNHN) de Paris, l’une des institutions majeures en recherche et patrimoine naturel dans le monde, est la cible d’une cyberattaque d’une ampleur inédite. L’organisation ne parvient plus à accéder à de nombreuses bases de données destinées à la recherche scientifique.
C’est une affaire qui s’enlise, et dont l’issue demeure incertaine.
Depuis plusieurs semaines, une partie des réseaux, des outils de recherche et des services numériques essentiels du Muséum National d’Histoire Naturelle de Paris restent inaccessibles.
L’incident, révélé le 31 juillet 2025 par nos confrères de La Tribune, n’a toujours pas été résolu à l’heure où nous publions cet article, ce mardi 12 août à la mi-journée.
La direction du Muséum dit faire face à une cyberattaque sévère : « C’est une attaque vraiment massive. (…) La durée de l’indisponibilité des outils et services, ainsi que le calendrier du retour à la normale, ne sont pour le moment pas encore déterminés », précise Gilles Bloch, président du MNHN, au micro de FranceInfo le 11 août 2025.
Pour l’heure, une question demeure : qui sont les auteurs de cette cyberattaque, et quelles peuvent être leurs motivations ?
L’hypothèse d’un ransomware
La direction de l’organisme confirme avoir prévenu les autorités. Une enquête judiciaire est en cours, dirigée par la section cybercriminalité du parquet de Paris, pour déterminer l’origine, le mode opératoire et les motivations exactes de l’attaque.
Si les premiers éléments semblent orienter vers une opération criminelle structurée, le cas du Muséum national d’Histoire naturelle va bien au-delà du simple vol de données, comme cela a pu être le cas lors de récentes cyberattaques ayant visé des grands groupes français tels qu’Air France ou Bouygues Telecom.
Ici, les chercheurs du Muséum et du centre PATRINAT se retrouvent privés d’accès à leurs principaux outils de travail. Les bases de données inaccessibles représentent une véritable manne scientifique, indispensable aux chercheurs et à plusieurs réseaux collaboratifs. L’attaque perturbe fortement la recherche française, particulièrement dans le secteur des sciences naturelles et de la biodiversité.
Et c’est précisément cette situation d’indisponibilité totale et d’interruption prolongée qui fait redouter la présence d’un ransomware. Il est probable que les auteurs de l’attaque cherchent à exercer un chantage financier : restaurer l’accès aux outils informatiques contre le versement d’une somme d’argent, le tout orchestré via un logiciel malveillant qui tient l’établissement en otage.
Une position claire de la part du MNHN
Dans sa communication publique, la direction du Muséum national d’histoire naturelle de Paris tient à lever toute ambiguïté : aucune rançon ne sera payée.
Gilles Bloch rappelle qu’il s’agit d’« une doctrine de l’État français et des administrations publiques ». L’objectif, comme dans d’autres pays, est de ne pas alimenter le modèle économique des réseaux cybercriminels.
En attendant l’issue de cette affaire, et malgré les perturbations techniques, l’établissement assure que les galeries d’exposition, les jardins botaniques et les parcs zoologiques restent ouverts et fonctionnent normalement. Les visiteurs ne subissent donc aucune conséquence directe de la cyberattaque.
canadianrecycler.ca - Toronto, Ontario -- Businesses across North America are reeling after a serious cyber attack threatened the data of 300 auto recycling businesses, including at least four based in Canada.
The attack, which occured on the evening of August 6, targeted businesses using SimpleHelp, a program that allows remote access to computer facilities. Those businesses that were caught up in the attack were locked out of their own databases and sent ransom notes demanding payment for the return of access.
Plazek Auto Recycling, near Hamilton, Ontario, was one of the businesses affected by the incident. According to Marc Plazek, employees only discovered the situation when they arrived at work to discover they were locked out of their computers — and discovered 30 copies of an identical ransom note on the printer.
“It was as if they arrived at our front gate, locked us in and said ‘we’ve got the only key.’ Except it was all done online.”
The ransomware software, LockBit Blpack, was developed by LockBit, a sophisticated cybercriminal organization. The group employs a dual-threat approach: it not only encrypt victims’ critical data and demand ransom payments for decryption keys, but also threaten to publicly leak sensitive information if its demands aren’t met – a tactic known as double extortion. First appearing on shadowy Russian forums in early 2020, LockBit has quickly established itself as a dominant force in the global ransomware landscape.
Like the other Canadian businesses affected by the hack, Plazek Auto Recycling did not respond to the threat. According to Marc Plazek, the company didn’t even entertain the idea of paying.
“We had a similar thing happen in 2019. We spoke with our insurance company who told us not to pay. They said there would be no reason for the hackers to bother living up to their word anyway.”
Because of the previous incident, Plazek Auto Recycling’s team had set up security measures and backed-up the computer system. The company was able to scrub its system of the malware and save all but a few hours worth of its records.
Other Canadian businesses known to have been affected include Millers Auto Recycling in Fort Erie, Ontario and Marks Parts in Ottawa. Fortunately, these companies were also able to restore access to data.
Other auto recyclers received assistance from the technical departments of Car-Part and Hollander. According to the Automotive Recyclers of Canada, most of the businesses affected by the attack had been
In response to the cyberattack, the executive director of the ARC, Wally Dingman, authored a column discussing the incident for this website.
hackernoon.com - Moonlock analysed Mac.c stealer, a new rival to AMOS. Learn its tactics, code reuse, and "building in public" strategy.
The story of the Mac.c stealer doesn’t begin with a major campaign or breach. It starts in the hushed corners of darknet forums, where a threat actor named 'mentalpositive' first emerged, drawing attention with a set of unusual traits that set him apart from other stealer developers.
Moonlock, the cybersecurity division of MacPaw, has been tracking mentalpositive for the past four months. We can already see that it is a new actor taking advantage of a macOS malware market that remains far less saturated than its Windows counterpart, marking the rise of the new wave of threat actors who are both technically skilled and commercially ambitious.
Although only recently active, Mac.c is already competing with larger, more established stealer operations like Atomic macOS Stealer. While it borrows heavily from AMOS and Rodrigo4 malware, it's tailored for quicker, high-impact data theft. As more URLs are added to its command-and-control infrastructure, Mac.c appears to be part of a larger underground ecosystem targeting macOS users.
What also stands out is a methodical and unusually transparent approach to building in public. 'mentalpositive' shared progress updates and even collected feedback on Mac.c builds — a surprising level of openness in the typically secretive world of macOS malware development.
In this article, we trace the evolution of Mac.c, unpack mentalpositive’s tactics, and examine how this stealer fits into the broader landscape of threats targeting Apple platforms.
A new player on the market
About four months ago, Moonlock Lab first noticed the emergence of the Mac.c stealer and attributed it to a developer under the alias 'mentalpositive'. This threat actor was one of many new players entering the macOS malware market, a space still far less crowded than the Windows-targeting malware industry.
Similar to other threat actors, 'mentalpositive' adopts recent trends in malware development: modular architecture for use across different campaigns, advanced obfuscation techniques, and increasingly complex command-and-control (C2) infrastructures.
However, the target profile and data exfiltration scope of mentalpositive’s Mac.c stand out. It harvests iCloud Keychain credentials, browser-stored passwords, crypto wallets, system metadata, and even files from specific locations on macOS — all using credentials obtained through phishing. By relying on standard system APIs and staged communication methods, it evades many traditional endpoint defences.
Building in public
Beyond technical design, 'mentalpositive' exhibited unusual behavior across darknet forums. Over the span of several months, this threat actor used one underground forum to showcase incremental updates to Mac.c, engage with potential users, and actively solicit feedback.
Such publicity may signal an intent to raise visibility and carve out a distinct market presence. It also appears to lay the groundwork for a custom stealer-as-a-service business model aimed squarely at the macOS threat niche.
The screenshots below show how the forum posts evolved over time as new features were announced. Since the original posts were written in Russian, we’ve included a brief explanation for each. The first screenshot shows an early advertisement offering a subscription to stealer updates for $1,500 per month.
edition.cnn.com | CNN Business - Millions of AT&T customers can file claims worth up to $7,500 in cash payments as part of a $177 million settlement related to data breaches in 2024.
The telecommunications company had faced a pair of data breaches, announced in March and July 2024, that were met with lawsuits.
Here’s a breakdown.
What happened?
On March 30, 2024, AT&T announced it was investigating a data leak that had occurred roughly two weeks prior. The breach had affected data until 2019, including Social Security numbers, and the information of 73 million former and current customers was found in a dataset on the dark web.
Four months later, the company blamed an “illegal download” on a third-party cloud platform that it learned about in April for a separate breach. This leak included telephone numbers of “nearly all” of AT&T cellular customers and customers of providers that used the AT&T network between May 1 and October 31, 2022, the company said.
The class-action settlement includes a $149 million cash fund for the first breach and a $28 million payout for the second breach.
Am I eligible for a claim?
AT&T customers whose data was involved in either breach, or both, will be eligible. Customers eligible to file a claim will receive an email notice, according to the settlement website.
AT&T said Kroll Settlement Administration is notifying current and former customers.
How do I file a claim?
The deadline to submit a claim is November 18. The final approval hearing for the settlement is December 3, according to the settlement website, and there could be appeals following an approval “and resolving them can take time.”
“Settlement Class Member Benefits will begin after the Settlement has obtained Court approval and the time for all appeals has expired,” the website states.
How much can I claim?
Customers impacted by the March incident are eligible for a cash payment of up to $5,000. Claims must include documentation of losses that happened in 2019 or later, and that are “fairly traceable” to the AT&T breach.
uk.news.yahoo.com - Records show hundreds of data breaches involving HMRC staff
HM Revenue and Customs (HMRC) has revealed that hundreds of staff have accessed the records of taxpayers without permission or breached security in other ways. HMRC dismissed 50 members of staff last year for accessing or risking the exposure of taxpayers’ records, according to The Telegraph.
354 tax employees have been disciplined for data security breaches since 2022, of whom 186 have been fired - and some were dismissed for accessing confidential information. HMRC holds sensitive data including salary and earnings, which staff cannot access without a good reason.
In an email to staff, the line manager of the claimant wrote: “There have been more incidents of this recently.”
John Hood, of accountants Moore Kingston Smith, said: “Any HMRC employee foolish enough to look up personal information that is not part of their usual responsibilities faces a ticking time bomb as most searches are tracked. As an additional security, some parts of the system are restricted so that only specifically authorised personnel can access them, such as the departments dealing with MPs and civil servants.”
HMRC’s annual report shows there were six incidents last year of employees changing customer records without permission, and two of staff losing inadequately protected devices.
A spokesman for HMRC said: “Instances of improper access are extremely rare, and we take firm action when it does happen, helping prevent a recurrence. We take the security of customers’ data extremely seriously and we have robust systems to ensure staff only access records when there is a legitimate business need.”
