The steady increase of digital scams and fraud Cybercrime on social media and messaging apps has forced major tech companies to significantly improve their security measures. It's becoming increasingly common for cybercriminals to attempt to steal accounts, bank details, or impersonate well-known brands through messages, ads, and contacts that appear harmless at first glance.
In this context, Meta is reinforcing its security strategy with a new generation of Anti-fraud AI applied to Facebook, WhatsApp and MessengerThe approach is clear: warn the user before they fall into the trap, minimize exposure to malicious content, and make life as difficult as possible for scammers operating on their platforms.
An anti-fraud AI to stay ahead of scams on Meta
Meta is deploying a set of features that combine artificial intelligence on the device itself and analysis on remote serversThis architecture allows, on the one hand, respecting end-to-end encryption in services like WhatsApp and Messenger, and on the other hand, offering an extra layer of advanced detection when the user decides to send certain suspicious content for review.
The tools rely on models capable of analyzing text, images, links, and behavior patternsThe goal is to identify everything from fraudulent ads that use celebrities' faces without permission to "technical support" messages that try to obtain verification codes, as well as friend requests from fake accounts that seek to gain the victim's trust.
According to data shared by the company, the problem is not insignificant: Meta claims to have disrupted millions of accounts and ads associated with organized fraud networks on Facebook, Instagram and WhatsApp, in some cases linked to real scam centers spread across several countries.
This technological offensive also relies on human review teams working side-by-side with AI, so that the models are trained with real examples and the margin of error in the classification of dubious content is reduced.
WhatsApp: firewall against account hijacking and suspicious groups
In the case of WhatsAppMeta's anti-fraud AI focuses on several particularly common attack vectors: account hijacking through device linking, technical support scams and deceptions in unknown groups.
One of the key new features is a system of Alerts when someone tries to link the account to an unrecognized device.If a suspicious request is detected, the app displays information such as the location from which the attempt originated, along with an explicit warning that it could be a scam. This allows the user to stop the process before a third party gains full access to their conversations.
AI also intervenes in situations of screen sharing during video calls with unknown contactsWhen this risk pattern is detected, the system issues an automatic warning, since many "tech support" scams rely on viewing the user's screen live to capture banking credentials, one-time codes, or other particularly sensitive information.
Furthermore, groups that users join without knowing much about their members now show a preventive safety panelThis panel offers details about the group, tips for detecting fraud attempts, and the option to leave without interacting, muting notifications until the user confirms they want to stay.
Meta is also testing Contextual alerts when starting chats with numbers that are not in the address bookIf the system identifies typical signs of phishing, fake investments, or unclear requests for money, it recommends exercising extreme caution and offers the possibility of blocking or reporting with a couple of taps.
Facebook: Scrutinizing friend requests and fake “support” pages
En FacebookAnti-fraud AI has two main fronts: suspicious friend requests and websites that impersonate customer service departments or established brands.
Upon receiving a friend request, the platform may display a warning if it detects risk factors: accounts with very few mutual friends, recently created profiles, geographic origin that doesn't match the user's network, or behaviors that match patterns associated with bot networks. This way, the user can better assess whether to accept, ignore, or block the request before opening the door to a potential fraudster.
Another relevant focus is the fake support pages and accounts that try to scam users seeking help to recover an account or resolve a problem. Meta claims to have taken action against tens of thousands of these sites, which typically requested login credentials, identity documents, or payments to "speed up" supposed recovery processes.
The company is also strengthening its mechanisms for Advertiser verification and fraudulent ad detectionThis includes those who use the images of celebrities, businesspeople, or public figures without permission. The AI compares faces and identifies recurring manipulations used to legitimize fake investments, nonexistent promotions, or products that never arrive.
This point is particularly sensitive for companies and content creators in Spain and the rest of Europe, where ads using using the image of entrepreneurs or well-known local figures to attract victimsgenerating confusion and reputational damage that is difficult to repair.
Messenger: User-configurable scam detection
The messaging app linked to Facebook, Messenger, incorporates a specific function called “Scam detection”This option, accessible through privacy and security settings, analyzes conversations initiated by unknown contacts in greater detail. Once activated, this feature provides a more in-depth analysis of conversations.
AI focuses on characteristic patterns of fake offersThese include: remote work offers demanding upfront payments, investments with guaranteed returns, urgent messages asking for money, or attempts to impersonate the support staff of well-known companies. If it detects anything unusual, it displays an alert indicating the type of risk present and suggests blocking or reporting the sender.
When expressly authorized by the user, the marked messages can be sent on a case-by-case basis to the Meta servers for deeper AI analysisThis step temporarily breaks end-to-end encryption only for that specific content, a decision that each person must weigh according to their risk tolerance and privacy concerns.
In parallel, Messenger has improved its system of link protectionThe Safe Browsing feature, which already analyzed URLs on the device itself, now relies on an additional layer called advanced protection, connected to constantly updated lists from dangerous sites. The design aims to block access to phishing pages or malware downloads without exposing the user's identity or full browsing history.
For support, sales, or customer service teams that use Messenger as their regular channel, these improvements represent both a An additional barrier against scams targeting your own brand as a somewhat safer environment for the users they interact with daily.
The artificial intelligence under the hood
Meta's new security features rely on a hybrid architecture that combines two complementary approaches. The first is the analysis on the deviceIt works directly on the user's mobile phone or computer and respects end-to-end encryption, as it does not need to send content to external servers to detect risk patterns.
This local analysis focuses on signals such as the fact that receive messages from unsaved numbersThese include attempts to share the screen with strangers, sudden changes in account behavior, or the appearance of links that match previous fraud reports. By operating on the device itself, the volume of data sent externally is reduced, providing an immediate first line of defense.
The second layer is activated when the user chooses to Send content for review using AI in the cloudIn this case, the flagged messages or items are processed by Meta's systems, which use more complex models to assess whether what is happening fits a known type of scam. If the analysis concludes that there is a risk, the platform displays explanations about the potential scam and recommends specific actions.
The company emphasizes that, in addition to AI, there are human moderation teams that review millions of pieces of content per dayThis human element is what allows, among other things, refining models, managing the most delicate cases, and dismantling large-scale coordinated campaigns that go beyond simple isolated messages.
This entire mechanism is complemented by agreements with security forces and other actors in the technology sector, in order to share intelligence on malicious domains, fake account networks and emerging tactics used by criminal organizations.
Main types of scams that Meta's AI is trying to stop
Meta platforms are a breeding ground for a wide variety of digital fraudsThe new tools are designed to detect and stop, among other things, the following types of common scams targeting users in Spain and Europe:
- Fake tech support scams: People who impersonate bank support services, operators, or even Meta itself to request verification codes, passwords, or remote access to the device.
- Screen sharing fraud: video calls in which the scammer convinces the victim to show their screen in order to capture bank details, SMS with one-time codes or credentials for services such as online banking or payment gateways.
- Non-existent job offers and scholarships: Advertisements or private messages promising easy or well-paid jobs in exchange for an initial payment for "procedures" or for providing sensitive personal information.
- Investment schemes and cryptocurrencies: campaigns that promise quick and guaranteed returns, often supported by the unauthorized use of famous people or well-known businessmen.
- Phishing and malicious links: URLs that mimic legitimate portals—banks, e-commerce platforms, public services—with the aim of stealing credentials or installing malware on the device.
The combination of automated monitoring, blacklists of dangerous sites, and user reporting mechanisms allows that The system adapts quickly to new variations of these scams.which often mutate to evade easier detection.
Impact on users, companies, and founders that depend on Meta
Beyond protecting the average user, the new AI anti-fraud features have direct effects on Businesses that use Facebook, WhatsApp, and Messenger as their basic infrastructure for sales, customer service, or community building.
On the one hand, if a company uses WhatsApp Business or Messenger to communicate with customers, Their messages could trigger alerts if the system perceives them as suspicious.For example, by including many links or requesting sensitive data. This necessitates reviewing scripts, tone, and frequency to minimize false positives and maintain audience trust.
On the other hand, startups in particularly sensitive sectors—such as fintech, digital health, or legal services— they find in these tools a clear sign: AI-based security is no longer an "extra" and is now part of the minimum expectations in any product or service that handles critical data.
Meta's growing collaboration with other technology players and public bodies also opens the door to New opportunities for companies specializing in cybersecurity, which can integrate their solutions with the threat detection ecosystems of large platforms or provide consulting services to smaller companies.
Ultimately, the effectiveness of these measures will depend on both users and organizations adopting good practices: activating available anti-fraud features, keeping apps updated, being wary of unsolicited messages, and reporting any suspicious behavior to feed detection systems.
With the implementation of its new Anti-fraud AI on Facebook, WhatsApp, and MessengerMeta is taking an important step towards a digital environment that is somewhat more difficult for scammers and a little safer for those who use its services daily, from individual users to companies that rely on these platforms for an essential part of their activity.
