forbes.com 20.08.2025 - xAI published conversations with Grok and made them searchable on Google, including a plan to assassinate Elon Musk and instructions for making fentanyl and bombs.
Elon Musk’s AI firm, xAI, has published the chat transcripts of hundreds of thousands of conversations between its chatbot Grok and the bot’s users — in many cases, without those users’ knowledge or permission.
Anytime a Grok user clicks the “share” button on one of their chats with the bot, a unique URL is created, allowing them to share the conversation via email, text message or other means. Unbeknownst to users, though, that unique URL is also made available to search engines, like Google, Bing and DuckDuckGo, making them searchable to anyone on the web. In other words, on Musk’s Grok, hitting the share button means that a conversation will be published on Grok’s website, without warning or a disclaimer to the user.
Today, a Google search for Grok chats shows that the search engine has indexed more than 370,000 user conversations with the bot. The shared pages revealed conversations between Grok users and the LLM that range from simple business tasks like writing tweets to generating images of a fictional terrorist attack in Kashmir and attempting to hack into a crypto wallet. Forbes reviewed conversations where users asked intimate questions about medicine and psychology; some even revealed the name, personal details and at least one password shared with the bot by a Grok user. Image files, spreadsheets and some text documents uploaded by users could also be accessed via the Grok shared page.
Among the indexed conversations were some initiated by British journalist Andrew Clifford, who used Grok to summarize the front pages of newspapers and compose tweets for his website Sentinel Current. Clifford told Forbes that he was unaware that clicking the share button would mean that his prompt would be discoverable on Google. “I would be a bit peeved but there was nothing on there that shouldn’t be there,” said Clifford, who has now switched to using Google’s Gemini AI.
Not all the conversations, though, were as benign as Clifford’s. Some were explicit, bigoted and violated xAI’s rules. The company prohibits use of its bot to “promot[e] critically harming human life or to “develop bioweapons, chemical weapons, or weapons of mass destruction,” but in published, shared conversations easily found via a Google search, Grok offered users instructions on how to make illicit drugs like fentanyl and methamphetamine, code a self-executing piece of malware and construct a bomb and methods of suicide. Grok also offered a detailed plan for the assassination of Elon Musk. Via the “share” function, the illicit instructions were then published on Grok’s website and indexed by Google.
xAI did not respond to a detailed request for comment.
xAI is not the only AI startup to have published users’ conversations with its chatbots. Earlier this month, users of OpenAI’s ChatGPT were alarmed to find that their conversations were appearing in Google search results, though the users had opted to make those conversations “discoverable” to others. But after outcry, the company quickly changed its policy. Calling the indexing “a short-lived experiment,” OpenAI chief information security officer Dane Stuckey said in a post on X that it would be discontinued because it “introduced too many opportunities for folks to accidentally share things they didn’t intend to.”
After OpenAI canned its share feature, Musk took a victory lap. Grok’s X account claimed at the time that it had no such sharing feature, and Musk tweeted in response, “Grok ftw” [for the win]. It’s unclear when Grok added the share feature, but X users have been warning since January that Grok conversations were being indexed by Google.
Some of the conversations asking Grok for instructions about how to manufacture drugs and bombs were likely initiated by security engineers, redteamers, or Trust & Safety professionals. But in at least a few cases, Grok’s sharing setting misled even professional AI researchers.
Nathan Lambert, a computational scientist at the Allen Institute for AI, used Grok to create a summary of his blog posts to share with his team. He was shocked to learn from Forbes that his Grok prompt and the AI’s response was indexed on Google. “I was surprised that Grok chats shared with my team were getting automatically indexed on Google, despite no warnings of it, especially after the recent flare-up with ChatGPT,” said the Seattle-based researcher.
Google allows website owners to choose when and how their content is indexed for search. “Publishers of these pages have full control over whether they are indexed,” said Google spokesperson Ned Adriance in a statement. Google itself previously allowed chats with its AI chatbot, Bard, to be indexed, but it removed them from search in 2023. Meta continues to allow its shared searches to be discoverable by search engines, Business Insider reported.
Opportunists are beginning to notice, and take advantage of, Grok’s published chats. On LinkedIn and the forum BlackHatWorld, marketers have discussed intentionally creating and sharing conversations with Grok to increase the prominence and name recognition of their businesses and products in Google search results. (It is unclear how effective these efforts would be.) Satish Kumar, CEO of SEO agency Pyrite Technologies, demonstrated to Forbes how one business had used Grok to manipulate results for a search of companies that will write your PhD dissertation for you.
“Every shared chat on Grok is fully indexable and searchable on Google,” he said. “People are actively using tactics to push these pages into Google’s index.”
ravenmail.io - Aug 14, 2025
In a recent credential phishing campaign, Raven AI (formerly Ravenmail) has uncovered attackers weaponizing Cisco's secure links to evade link scannin.
Picture this: You receive an email with a link that starts with "secure-web.cisco.com" Your brain immediately registers "secure" and "Cisco" – two words that scream safety and reliability. You click without hesitation. After all, if Cisco is protecting the link, it must be safe, right?
Unfortunately, cybercriminals are banking on exactly that assumption – and traditional email security solutions are falling for it too. But Raven's context-aware AI recently caught a sophisticated attack that perfectly illustrates how attackers weaponize trusted security infrastructure.
The Irony of Trust
Cisco Safe Links represents one of cybersecurity's most elegant solutions – and its most exploitable weakness. Designed as part of Cisco's Secure Email Gateway and Web Security suite, Safe Links works by rewriting suspicious URLs in emails, routing clicks through Cisco's scanning infrastructure before allowing users to reach their destination. Think of it as a digital bodyguard that checks every door before you walk through it.
The technology mirrors similar offerings from Microsoft Defender and Proofpoint TAP. When you click a protected link, Cisco's systems perform real-time threat analysis, blocking malicious destinations and allowing legitimate ones. It's a brilliant concept that has undoubtedly prevented countless successful phishing attacks.
But here's where the story takes a dark turn: attackers have figured out how to turn this protective mechanism into their own weapon.
The Attack Vector That Shouldn't Exist
The scheme is diabolically simple. Cybercriminals deliberately embed legitimate Cisco Safe Links into their phishing campaigns, creating a perfect storm of misdirected trust. Here's why this approach is so devastatingly effective:
Trust by Association: When users see "secure-web.cisco.com" in a URL, they instinctively assume it's been vetted and approved. The Cisco brand carries enormous weight in cybersecurity circles – seeing it in a link feels like getting a security clearance stamp.
Bypass Detection Systems: Many email security gateways focus their analysis on the visible domain in URLs. When that domain is "secure-web.cisco.com", it often sails through filters that would otherwise flag suspicious links.
The Time Gap Advantage: Even Cisco's robust threat intelligence needs time to identify and classify new threats. Attackers exploit this window, using freshly compromised websites or newly registered domains that haven't yet been flagged as malicious.
How Attackers Generate Cisco's Links
You might wonder: how do cybercriminals get their hands on legitimate Cisco Safe Links in the first place? The methods are surprisingly straightforward:
Method 1: The Inside Job
Attackers compromise or create accounts within Cisco-protected organizations. They simply email themselves malicious links, let Cisco's system rewrite them into Safe Links, then harvest these URLs for their campaigns.
Method 2: The Trojan Horse
Using compromised email accounts within Cisco-protected companies, attackers send themselves test emails containing malicious links. The organization's own security infrastructure helpfully converts these into trusted Safe Links.
Method 3: The SaaS Backdoor
Many cloud services send emails through Cisco-protected environments. Attackers sign up for these services, trigger automated emails to themselves containing their malicious links, and receive back the Cisco-wrapped versions.
Method 4: The Recycling Program
Sometimes the simplest approach works best. Attackers scour previous phishing campaigns for still-active Cisco Safe Links and reuse them in new attacks.
Raven AI Catches the Attack in Action
Recently, RavenMail's context-aware AI detected a perfect example of this attack technique in the wild. The phishing email appeared legitimate at first glance – a professional-looking "Document Review Request" from what seemed to be an e-signature service.
This is an AI-overview of the attack, this is not just the summary of the attack but the detection engine has context of the organization and consumes relevant signals to make a verdict.
Raven AI in action
Here's what made this attack particularly sophisticated:
The Setup: The email claimed to be from "e-Sign-Service" with a Swiss domain, requesting document review for a "2025_Remittance_Adjustment" file. Everything looked professional – proper branding, business terminology, and a clear call-to-action.
The Cisco Safe Links Component: While this particular example shows the final malicious URL, the attack pattern follows the exact methodology we described – using trusted domains and legitimate-looking parameters to bypass detection systems.
What RavenAI Spotted: Unlike traditional email security solutions that might have been fooled by the professional appearance and trusted domain elements, RavenMail's context-aware AI identified several red flags:
Inconsistent sender identity (e-signature service from a non-standard domain)
Suspicious URL structure with encoded parameters
Document request patterns commonly used in credential phishing
Contextual anomalies in the business process workflow
The smoking gun? This wasn't a random phishing attempt – it was a carefully crafted attack designed to exploit user trust in legitimate business processes and security infrastructure.
Why Traditional Security Missed This
This attack would likely have bypassed many conventional email security solutions for several reasons:
Professional Appearance: The email looked like a legitimate business communication – complete with proper formatting, business terminology, and what appeared to be a standard document review workflow.
Domain Trust: While not using Cisco Safe Links directly, the attack employed similar trust-exploitation tactics by using a domain structure that appeared legitimate.
Context Deception: The attack leveraged realistic business scenarios (document review, remittance adjustments) that users encounter daily in professional environments.
Multi-Layer Misdirection: By providing both a primary button and an "alternative access method," the attacker created multiple attack vectors while appearing helpful and legitimate.
The Raven AI Advantage: Context-Aware AI Detection
Context-aware artificial intelligence that goes beyond simple domain and signature-based detection:
Business Process Understanding: Raven's AI understands legitimate business workflows and can identify when communications deviate from expected patterns – even when they look professionally crafted.
Multi-Signal Analysis: Rather than relying solely on domain reputation or static signatures, the AI analyzes multiple contextual signals simultaneously to identify sophisticated attacks.
Behavioral Pattern Recognition: The system recognizes common attack methodologies, including trust exploitation tactics that leverage legitimate-seeming domains and professional formatting.
Real-Time Adaptation: As attackers evolve their techniques, RavenMail's AI continuously learns and adapts, staying ahead of emerging threats like Safe
The Bigger Picture: Why Context-Aware AI Matters
This detection illustrates a fundamental shift in cybersecurity: attackers are no longer just exploiting technical vulnerabilities – they're weaponizing human psychology and business processes.
This isn't just about Cisco Safe Links abuse (though that remains a significant threat). It's about a new class of attacks that exploit our trust in legitimate business processes, professional communication patterns, and security infrastructure itself.
Traditional signature-based and reputation-based security solutions struggle with these attacks because they look legitimate at every technical level. The malicious elements are hidden in context, behavior, and the subtle exploitation of trust relationships.
Context Over Content: Rather than just analyzing what's in an email, RavenMail's AI understands what the email is trying to accomplish and whether that aligns with legitimate business processes.
Trust Verification: The system doesn't just trust professional appearance or legitimate-looking domains – it actively verifies the contextual appropriateness of communications.
Adaptive Learning: As attackers develop new trust exploitation techniques (like Safe Links abuse), AI-driven solutions can adapt without requiring manual rule updates.
Proactive Defense: Instead of waiting for attacks to succeed and then updating blacklists, context-aware AI can identify attack patterns before they cause damage.
The most effective defense against modern email threats isn't just about blocking bad domains or scanning attachments – it's about understanding the attacker's intent and recognizing when legitimate-looking communications serve malicious purposes
ncsc.admin.ch 19.08.2025 - La semaine dernière, deux incidents ont été signalés à l’OFCS dans lesquels des cybercriminels ont tenté d’inciter des destinataires à installer des logiciels malveillants. Ces cas illustrent à quel point les méthodes ont évolué : Il est aujourd’hui beaucoup plus difficile d’introduire un logiciel malveillant sur un ordinateur qu’il y a quelques années. Les cybercriminels ont donc de plus en plus recours à des techniques sophistiquées d’ingénierie sociale pour atteindre leurs objectifs.
Au cours du premier semestre 2025, seuls 182 cas liés à des logiciels malveillants ont été signalés à l’OFCS via le formulaire de signalement, ce qui ne représente qu’environ 0,4 % de l’ensemble des signalements reçus pendant cette période. Ce faible nombre peut être interprété de deux manières. D’un point de vue positif, les mécanismes de protection techniques tels que les programmes antivirus et les filtres anti-spam sont désormais si efficaces qu’ils bloquent la plupart des attaques de logiciels malveillants avant même qu’elles ne soient exécutées, ce qui réduit le nombre de signalements à l’OFCS. Dans une optique négative, on peut toutefois considérer que les attaques sont devenues si sophistiquées qu’elles ne sont pas détectées par les personnes concernées et ne font donc pas l’objet de signalements.
Le recours à des logiciels malveillants n’a toutefois pas complètement disparu. Ces derniers temps, l’OFCS reçoit à nouveau davantage de signalements concernant des e-mails visant à diffuser des logiciels malveillants, comme l’illustrent les deux exemples actuels suivants, survenus la semaine dernière.
Fausse facture au nom d’Intrum
La semaine dernière, de fausses factures ou des rappels ont été envoyés par e-mail au nom de la société de recouvrement Intrum. Le message fait référence à une prétendue facture QR en pièce jointe, qui doit être ouverte pour être payée. En réalité, la pièce jointe n’est pas un fichier PDF, mais un fichier HTML.
Lors de l’ouverture, le destinataire reçoit un message indiquant que le fichier PDF ne peut pas être affiché, car JavaScript est désactivé.
Pour activer le JavaScript, il faut appuyer sur les touches « Windows+R » et « Ctrl+V » – une méthode bien connue qui a déjà été évoquée dans une précédente rétrospective hebdomadaire de l’OFCS. Lors du chargement du fichier HTML, un script PowerShell malveillant est copié dans le presse-papiers de l’ordinateur. Un script PowerShell est un fichier texte contenant les commandes que l’ordinateur doit exécuter.
En appuyant sur cette combinaison de touches, une fenêtre s’ouvre dans laquelle vous pouvez exécuter des commandes. La combinaison de touches « Ctrl+V » insère la commande malveillante du presse-papiers dans cette fenêtre. L’ordinateur se connecte ensuite à un serveur des malfaiteurs et un logiciel malveillant est téléchargé et installé.
Prétendue demande de paiement de l’UBS
Un autre cas concerne une fausse demande de paiement au nom d’UBS. Là encore, le destinataire est invité à ouvrir un fichier joint à l’e-mail. Dans ce cas, il s’agit bien d’un fichier PDF, mais celui-ci est protégé par un mot de passe. Le mot de passe est toutefois fourni directement dans l’e-mail, pour plus de commodité.
Cette procédure sert probablement à contourner les mécanismes de sécurité tels que les filtres anti-spam et les programmes antivirus, car ceux-ci ne peuvent pas analyser le contenu du fichier protégé par mot de passe. Une fois le mot de passe saisi, le fichier PDF s’ouvre et indique que le véritable contenu se trouve sur un lecteur OneDrive.
Le lien indiqué mène au téléchargement d’un fichier archive contenant un fichier « batch ». Il s’agit d’un fichier texte exécutable qui contient des commandes. Son exécution entraîne le téléchargement et l’installation du logiciel malveillant, comme dans le premier exemple.
Ces deux exemples montrent à quel point les attaques sont désormais complexes et articulées sur plusieurs niveaux. Un simple fichier exécutable joint à un e-mail ne suffit plus depuis longtemps pour infecter un ordinateur. Les pirates misent désormais sur des manœuvres de diversion sophistiquées pour contourner les mécanismes de protection techniques et inciter les victimes à jouer un rôle actif. La vigilance et la sensibilisation restent donc des éléments centraux de la cybersécurité.
Recommandations
Ne cliquez pas sur les liens contenus dans les e-mails et les SMS.
Si vous attendez un rappel, contactez l’agence de recouvrement ou la caisse d’assurance maladie afin de vérifier si la créance est bien justifiée. Utilisez pour cela les coordonnées figurant sur les sites Internet officiels des entreprises.
Si vous soupçonnez l’installation d’un logiciel malveillant, adressez-vous à un magasin spécialisé en informatique. Le plus sûr est de réinstaller complètement votre ordinateur. N’oubliez pas de sauvegarder toutes vos données personnelles au préalable.
Après la réinstallation, modifiez tous vos mots de passe pour tous vos accès en ligne (e-mail, réseaux sociaux, etc.).
blog.pypi.org - The Python Package Index Blog - PyPI now checks for expired domains to prevent domain resurrection attacks, a type of supply-chain attack where someone buys an expired domain and uses it to take over PyPI accounts through password resets.
These changes improve PyPI's overall account security posture, making it harder for attackers to exploit expired domain names to gain unauthorized access to accounts.
Since early June 2025, PyPI has unverified over 1,800 email addresses when their associated domains entered expiration phases. This isn't a perfect solution, but it closes off a significant attack vector where the majority of interactions would appear completely legitimate.
Background
PyPI user accounts are linked to email addresses. Email addresses are tied to domain names; domain names can expire if unpaid, and someone else can purchase them.
During PyPI account registration, users are required to verify their email addresses by clicking a link sent to the email address provided during registration. This verification ensures the address is valid and accessible to the user, and may be used to send important account-related information, such as password reset requests, or for PyPI Admins to use to contact the user.
PyPI considers the account holder's initially verified email address a strong indicator of account ownership. Coupled with a form of Two-Factor Authentication (2FA), this helps to further secure the account.
Once expired, an attacker could register the expired domain, set up an email server, issue a password reset request, and gain access to accounts associated with that domain name.
Accounts with any activity after January 1 2024 will have 2FA enabled, and an attacker would need to have either the second factor, or perform a full account recovery.
For older accounts prior to the 2FA requirement date, having an email address domain expire could lead to account takeover, which is what we're attempting to prevent, as well as minimize potential exposure if an email domain does expire and change hands, regardless of whether the account has 2FA enabled.
This is not an imaginary attack - this has happened at least once for a PyPI project back in 2022, and other package ecosystems.
TL;DR: If a domain expires, don't consider email addresses associated with it verified any more.
itnews.com.au - TPG Telecom has revealed that iiNet’s order management system was breached by an unknown attacker who abused legitimate credentials to gain access.
The telco said [pdf] that it “appears” that a list of email addresses and phone numbers was extracted from the system.
“Based on current analysis, the list contained around 280,000 active iiNet email addresses and around 20,000 active iiNet landline phone numbers, plus inactive email addresses and numbers,” TPG said.
“In addition, around 10,000 iiNet usernames, street addresses and phone numbers and around 1700 modem set-up passwords, appear to have been accessed.”
The order management system is used to create and track orders for iiNet services.
TPG Telecom said that the system does not store “copies or details of identity documents, credit card or banking information.”
The telco apologised “unreservedly” for the incident and said it would contact all iiNet customers, both those impacted as well as “all non-impacted iiNet customers to confirm they have not been affected.”
Investigations so far have not uncovered any escalation of the breach by the attacker beyond the order management system.
TPG Telecom has advised relevant government agencies of the incident.
WARSAW, Aug 14 (Reuters) - A large Polish city could have had its water supply cut off on Wednesday as a result of a cyberattack, a deputy prime minister said after the intrusion was foiled.
In an interview with news portal Onet on Thursday, Deputy Prime Minister Krzysztof Gawkowski, who is also digital affairs minister, did not specify who was behind the attack or which city was targeted.
Poland has said that its role as a hub for aid to Ukraine makes it a target for Russian cyberattacks and acts of sabotage. Gawkowski has described Poland in the past as the "main target" for Russia among NATO countries.
Gawkowski told Onet that the cyberattack could have meant there would be no water in one of Poland's big cities.
"At the last moment we managed to see to it that when the attack began, our services had found out about it and we shut everything down. We managed to prevent the attack."
He said Poland manages to thwart 99% of cyberattacks.
Gawkowski last year that Poland would spend over 3 billion zlotys ($800 million) to boost cybersecurity after the state news agency PAP was hit by what authorities said was likely to have been a Russian cyberattack.
The digital affairs ministry did not immediately respond to an email requesting further details.
On Wednesday Prime Minister Donald Tusk, who has warned that Russia is trying to drive a wedge between Warsaw and Kyiv, said that a young Ukrainian man had been detained for acts of sabotage on behalf of foreign intelligence services, including writing graffiti insulting Poles.
PAP reported on Thursday that a 17-year-old Ukrainian man detained, among other things, for desecrating a monument to Poles killed by Ukrainian nationalists in World War Two has been charged with participating in an organised criminal group aimed at committing crimes against Poland.
nist.gov - Face morphing software, which combines photos of different people into a single image, is being used to commit identity fraud
August 18, 2025
bleepingcomputer.com - Plex has notified some of its users on Thursday to urgently update their media servers due to a recently patched security vulnerability.
The company has yet to assign a CVE-ID to track the flaw and didn't provide additional details regarding the patch, only saying that it impacts Plex Media Server versions 1.41.7.x to 1.42.0.x.
Yesterday, four days after releasing security updates that addressed the mysterious security bug, Plex emailed those running affected versions to update their software as soon as possible.
"We recently received a report via our bug bounty program that there was a potential security issue affecting Plex Media Server versions 1.41.7.x to 1.42.0.x. Thanks to that user, we were able to address the issue, release an updated version of the server, and continue to improve our security and defenses," the company said in the email.
"You're receiving this notice because our information indicates that a Plex Media Server owned by your Plex account is running an older version of the server. We strongly recommend that everyone update their Plex Media Server to the most recent version as soon as possible, if you have not already done so."
Plex Media Server 1.42.1.10060, the version that patches this vulnerability, can be downloaded from the server management page or the official downloads page.
While Plex hasn't shared any details regarding the vulnerability so far, users are advised to follow the company's advice and patch their software before threat actors reverse engineer the patches and develop an exploit.
Although Plex has experienced its share of critical and high-severity security flaws over the years, this is one of the few instances where the company has emailed customers about securing their systems against a specific vulnerability.
In March 2023, CISA tagged a three-year-old remote code execution (RCE) flaw (CVE-2020-5741) in the Plex Media Server as actively exploited in attacks. As Plex explained two years earlier, when it released patches, successful exploitation can allow attackers to make the server execute malicious code.
While the cybersecurity agency didn't provide any information on the attacks exploiting CVE-2020-5741, they were likely linked to LastPass' disclosure that one of its senior DevOps engineers' computers had been hacked in 2022 to install a keylogger by abusing a third-party media software RCE bug.
The attackers exploited this access to steal the engineer's credentials and compromise the LastPass corporate vault, resulting in a massive data breach in August 2022 after stealing LastPass's production backups and critical database backups.
The same month, Plex also notified users of a data breach and asked them to reset passwords after an attacker gained access to a database containing emails, usernames, and encrypted passwords.
euractiv.com - MADRID – Spanish magistrates, law enforcement leaders and opposition politicians are voicing alarm over Madrid’s unusually close ties to Beijing, as the Chinese tech giant’s footprint in Spain’s public sector is deeper than first thought.
The concerns have intensified since July, when reports surfaced of an alleged €12.3 million contract between 2021 and 2025 for Huawei to store sensitive judicial wiretap data for the interior ministry.
Opposition Popular Party (PP) secretary general Miguel Tellado branded the public tender “shady” and claimed it was part of “the Chinese branch of Pedro Sánchez’s enormous corruption network.” The PP is also demanding that Sánchez’s top ministers testify before parliament after the summer recess.
The interior ministry has denied the existence of the Huawei agreement and did not clarify whether the initial €12.3 million figure was part of a broader deal with Spanish firms such as Telefónica, TRC or Econocom, as several local outlets have suggested.
The alleged deal has landed at a politically delicate moment for the Socialist-led government, already reeling from multiple corruption scandals.
