I asked a cybersecurity expert how to shop more safely with AI — these are their 5 non-negotiables

(Image credit: Dr. Serena Sullivan/Shutterstock) We’ve gotten to the point where people are relying on AI assistants to help them navigate all the madness associated with holiday shopping.According to Bain & Company, nearly a quarter of US online shoppers plan to kick off their holiday shopping on an AI platform like ChatGPT or Gemini (which is a stat that’s up from 17% last year).With that increase in AI tools being used to aid online shoppers comes the threat of scammers, who look to exploit those shoppers with fake online stores (which are now popping up at a current rate of one every three minutes). And what’s worse is that those same stores have now started showing up in AI assistants’ product recommendations.In a bid to learn more about how to shop more safely when using AI for online shopping, how a holiday gift list given to AI could build out an identity map of your family and more, I spoke to cybersecurity expert Serena Sullivan.Here’s hoping her advice makes your AI-assisted holiday shopping experience less susceptible to any sorts of scams. (Image credit: Shutterstock)When someone asks an AI assistant to help with their holiday shopping, what personal information might they be revealing about themselves without realizing it?Consumers often unintentionally disclose personal information. Depending on the way the user crafts their prompts into the AI engine, some information could be inferred. For example, if the user is searching for personal care items to purchase as a gift, they could unintentionally reveal the gender and age of the person they are purchasing the gift for. If they search for ballet shoes, they could reveal the age range of the recipient, the gender, their hobbies and interests. If you are looking for gifts that are very expensive versus inexpensive, some conclusions could be drawn about the user’s disposable income. If the person was looking for an experience instead of a gift, maybe a family trip to a theme park, AI could likely deduce the size of the family, whether any family members need accessibility options, which state or country they live in, and a generalization of how much income the family has access to.Individually, these details might seem harmless. However, when combined, they can create a surprisingly detailed picture of someone's personal life. This is how a data aggregation attack works. An attacker can pull information using Open Source Intelligence (OSINT), meaning information that is freely available on the internet, to compile a detailed profile of the individual. This information can then be used to perform detailed phishing or other social engineering attacks through email, voicemail, phone, or text.Treat AI assistants as you would any other online service. Only share the information you are willing to disclose, keeping in mind that other information could be combined to create a detailed picture of the user and those they are purchasing gifts for.Treat AI assistants as you would any other online service. Only share the information you are willing to disclose, keeping in mind that other information could be combined to create a detailed picture of the user and those they are purchasing gifts for. If an AI shopping assistant recommends a website that looks real, how can shoppers tell whether it’s actually a scam or not? And what new warning signs should they keep an eye out for?Just because AI makes a recommendation does not mean that AI found the correct website and does not indicate that the website has been vetted for safety or security. Additionally, AI could present the shopper with an outdated link.When you click a link provided in AI results, the first thing to check is if it is a secure website. For example, if the shopper is warned that the digital certificate is out of date or illegitimate, do not continue using that website. Other warning signs, beyond security, are if the items appear too good to be true. For example, if you know an item normally costs about $100 and this website is selling it for $10, it could be a scam.Shoppers can also identify potential signs of scams by checking the website address and determining if it matches what was expected. For example, the shopper wants to buy a nice piece of jewelry for their significant other; let’s say they wanted to go to ringsforspecialpeople.com (please note this is a fake website, just for the example). When the shopper enters the website, they notice the domain name is not related; maybe it automatically turns into jewelryforspecialpeople.com. The website could be an exact duplicate of the website the shopper is accessing, but it is not. This should not be confused with related domains. For example, many companies will purchase all the related domains and have them redirected to their main website. The owner of the website could have purchased jewelryforspecialpeople.com and have it redirected to the main website.Shoppers can also navigate to the suggested website without clicking the link in the AI response. Instead of clicking the link, the shopper could use a search engine to search for that website and click the website from the search results.Prospective buyers can also review the payment options the website provides. One red flag is if the website requests payment in the form of money orders, wire transfers, gift cards, or cryptocurrency. Those forms of payment are not typical. Most companies would prefer a credit card or payments through other sites, like PayPal, Apple Pay, and others.Shoppers should be aware of the company they are doing business with. They can research the seller by verifying it is a legitimate company, read user reviews, and identify if they have a customer service department. Most scammers will not have a customer service department. Review the policies on the website. Another red flag is if the privacy policy is non-existent or poorly written. Legitimate companies will have a privacy policy, written by lawyers, using legal jargon. If the policy does not appear to be written by a lawyer, the shopper should not spend their money there.The warning signs mentioned above apply regardless of how someone arrives at a website. Whether a shopper clicks a link provided by an AI assistant or finds the website through a traditional search engine, the same precautions should be taken. However, as AI increasingly generates recommendations based on individual shopping queries, fraudulent websites and scams may become more difficult to recognize, particularly when the results appear relevant, personalized, and legitimate.The advancement of AI has given scammers the ability to create highly convincing websites, customer reviews, product descriptions, and chatbots. As a result, it can be increasingly difficult for shoppers to distinguish legitimate businesses from fraudulent ones. However, consumers can take steps to verify a business before making a purchase. Researching the company, reviewing its Better Business Bureau (BBB) rating and complaint history, and checking for any current or past consumer protection lawsuits or regulatory actions can help shoppers determine whether a business is trustworthy.If a website appears to be professional-looking, it does not necessarily indicate that the website is trustworthy. If a shopper has even the slightest inkling that the website may be fraudulent, they should research using some of these suggestions or find a different website that does not contain any of the red flags mentioned.Just because AI makes a recommendation does not mean that AI found the correct website and does not indicate that the website has been vetted for safety or security.Which popular holiday tech gifts pose the biggest privacy risks? And what should parents check before buying them?Children would often like to be gifted technology items. However, parents should use caution when deciding what type of technology is appropriate for their child. I would stress that parents teach their children how to spot behaviors or platforms that might be attempting to collect information, or worse. Many free K-12 educational resources are available to help parents create safer online experiences for their children. Parents should evaluate the technology their children use, taking into consideration privacy, security, age-appropriateness, and individual maturity levels. I often encourage parents to seek out educational materials that teach children about online safety, including how to recognize potentially harmful situations and what steps to take if they encounter something concerning online.Governance exists to offer protection for children under 13. One of the most important pieces of legislation is the Children's Online Privacy Protection Act (COPPA). Companies covered by COPPA must meet specific requirements when collecting personal information from children. While this governance is in place to provide guidance to companies producing the technology, the first line of defense is the child and their parents.Many popular technology gifts have the capability of collecting personal information. Parents are the first line of defense for minimizing some of the risks associated with those types of products. If the gift includes options for using a microphone, cameras, location tracking, or has AI integrated, parents should pay particular attention to those capabilities and become familiar with parental controls.AI-driven toys, like robots or interactive dolls, could record the child’s voice, save conversations and questions the child may ask. Other interactions could also be saved on the device, or in the cloud if it is connected to the internet. If the toy connects to the internet, there is potential that an attacker could access that information.Video games are a very common gift for children to receive. Games such as Roblox have a feature where children can talk and type to other users they are playing with. A safer approach is to use parental controls to block that feature from children.If children are using personal assistants or planners and add significant dates, such as birthdays, vacations, school hours, extracurricular activities, and similar, that could be dangerous if an attacker compromises that data. The attacker would potentially be privy to household schedules, how many members are in the family, potentially gender, sexual orientation of family members, religious beliefs, or when the child would be away from home and where they are.Smartwatches and fitness trackers can also pose potential danger. These devices may collect location information, activity patterns, contacts, and other personal information. Many of these devices are not covered under HIPAA, so they do not necessarily have the same protection applied as medical devices do.Parents can ask themselves a few questions when evaluating the potential technology gift:1. What information does the device collect?2. Is audio, video, or location information stored in the cloud?3. Who has access to that information, and is it shared with third parties?4. Can parents disable microphones, cameras, and location tracking?5. Does the manufacturer provide regular security updates?6. Can parents review and permanently delete collected information?7. Does the manufacturer have a history of security vulnerabilities or privacy complaints? (Image credit: Shutterstock)A lot of people use AI to manage their holiday lists, plan family gatherings and find the best gifts for their loved ones. At what point does sharing those details become a privacy risk? And what information should never be entered into an AI assistant in those cases?Users should be careful about what they self-disclose. Sharing general information about a family gathering, planning a holiday list, or finding the best gifts for their loved ones is a reasonable use of technology. However, one does not need to include things like full names, addresses, phone numbers, birth date, financial circumstances, sexual orientation, ethnicity, and other identifiable information. Users should consider sharing the bare minimum of information with the AI assistant that will allow it to perform as expected.Certain information should never be provided to any AI-enabled device, such as:Personally identifiable information, for adults and children, like Social Security numbers, driver's license numbers, passport information, school schedules, extracurricular activities, or information that could allow an attacker to know where the user is physically located.As a general rule of thumb, if you would not want anyone outside of your family and close friends to know the information, it should not be entered into the technology.Why might grandparents and other less tech-savvy online shoppers be especially vulnerable to AI-assisted shopping scams? How can their families help protect them?Anyone who is not familiar with technology or does not understand online safety practices can be especially vulnerable to scammers. Older adults are often targeted because the scammers know their victims are unfamiliar with online safety and security measures and will use terminology to increase the likelihood that an inexperienced user might fall for the scam. Older adults make desirable victims because they may have more funds, like access to a retirement account, which could contain a large sum intended to last the retiree the rest of their life. Scammers will often try to create a sense of urgency, exploit trust, or play on emotions.Scammers used to craft emails, websites, apps, and other methods of delivery, and they were often riddled with misspellings, logos that did not seem accurate, and incorrect grammar. With generative AI, scammers have reduced the likelihood of consumers immediately identifying the communication as a scam. By using AI and a few pointed prompts, scammers can now craft grammatically correct and spelling error-free communication. AI has made it much easier for everyone to develop websites and apps in a fraction of the time it used to take.Creating more realistic content, combined with the ability to find publicly available information about their victim, has allowed scammers to develop realistic-looking communications that can be specifically targeted to their victim. For example, Dr. Sullivan could have posted to an open Facebook account that she likes to take her corgi to the dog park. The attacker could create an email pretending to be from the city that owns the dog park, naming her dog, saying he was being banned because someone reported her dog as being aggressive. They could include an urgent deadline to have the situation rectified, increasing the chances that Dr. Sullivan would click on the link in the email. When she clicks the link, it could take her to the fake dog park’s website that appears to be legitimate. Perhaps they will encourage her to pay a fine if she wants her dog to be allowed back in the park. Just that quickly, the attacker could now have a username and password she created on the website, along with her credit card information. That information could then be used to steal additional funds from her or start trying the username and password she provided to attempt to enter other accounts and websites. Voice cloning is also easier to perform. All the attacker needs is a few-second clip of the voice, and they can generate illegitimate phone calls and other types of voice communications. Imagine a grandparent receiving a phone call in their grandchild’s voice, stating they are being held for ransom. The fake voice states that the grandparents must send a certain amount of cryptocurrency to see their grandchild again. These situations have happened, and some have been tricked, while others knew it was not their grandchild, because the grandchild is sitting in the back seat of their car and could not be the same person on the telephone. If a familiar voice is saying unusual things, it is a great idea to double-check to determine if it was a legitimate phone call or not.Families can protect themselves by using several precautions.Train all family members in common scams and indicators that could lead to communication being a scam.Encourage them to shop directly through established retailers rather than clicking advertisements or unfamiliar links.Help them install automatic updates and use multifactor authentication.Set up transaction alerts on credit cards and bank accounts.Encourage them to call a family member directly if they receive an unexpected emergency request.Tips on how to handle online scams are to slow down and think about the request, who is making the request, and how likely the request is to be legitimate. Scammers thrive on people making hasty decisions or acting out of emotions rather than thinking the situation through. From the ransom situation described above to a fantastic and unbelievable deal on an item the person was researching as a gift, consumers should take a few minutes to ask for help from a tech-savvy family member or to think about the situation thoroughly before acting. (Image credit: Shutterstock)What privacy settings or safeguards should online shoppers turn on before using ChatGPT, Gemini or any other AI assistant to research gifts, compare prices or make purchases?Here are a few suggestions on how to use AI in a safer manner. So many AI platforms exist, so it would be impossible to list all of them here; only some of the most popular AI is listed here. If a person is using a different AI platform outside of what is listed here, users can perform a search to determine how to perform these actions on other platforms. Consumers should also be aware that even with these pointers and training, nothing is 100% secure.ChatGPTAccess ChatGPT’s settings and select Data Controls, then toggle off the option to allow “Improve the Model for Everyone.”Use Temporary Chat when you do not want a conversation saved in your regular chat history.Review the personalization settings to understand how memory is being used.Review connected apps and third-party permissions before authorizing an AI assistant to access other services.Turning off model improvement does not automatically delete existing conversations, and temporary chats may also be retained for up to 30 days for safety purposes.For specific details, conduct a search for ChatGPT Data Controls.Google GeminiReview Gemini Keep Activity settings.Turn off Keep Activity if you do not want your activity saved and used under the associated settings.Review the automatic deletion period for stored activity.Review connected applications and the information they can access.If the consumer is going to upload documents, photos, files and other media, be careful of what is shared. Erase metadata before you upload it, pay attention to what is in the background of a photo being shared, and do not upload sensitive information.A consumer can perform a search for Gemini Apps Privacy Hub to learn more details. (Image credit: Shutterstock)If you could give everyone a five-point safety checklist for using AI to shop this holiday season, what would be on it? And what should they always verify themselves before clicking “buy”?1. Protect your personal information: Do not share passwords, financial information, Social Security numbers, children's identifying information, or detailed travel schedules with AI assistants.2. Verify every website and seller: Never assume an AI-recommended website is legitimate. Independently verify the retailer, domain name, seller reputation, and contact information before purchasing.3. If a deal seems too good to be true, it probably isn’t true: Be cautious of unusually low prices, unexpected discounts, fast-approaching sales end, and requests for unusual payment methods. If something seems urgent, that is an indicator that consumers should slow down and independently verify the details.4. Secure your accounts and devices: Use multifactor authentication when possible, do not use the same password for every website, install security updates, and review privacy settings before connecting shopping accounts or services to AI assistants.5. Verify before you buy: Independently confirm the product, seller, total price, shipping costs, return policy, and payment method.Details a consumer should verify before purchasing include ensuring the seller is legitimate, validating the details of the product being purchased, confirming the total cost, payment security, and available purchase protections. Use credit cards for purchases whenever possible, as that allows consumers one more avenue to explore should they have followed all the safety tips and were scammed regardless. Follow Tom's Guide on Google News and add us as a preferred source to get our up-to-date news, analysis, and reviews in your feeds. Subscribe to Tom's Guide on YouTube and follow us on TikTok.More from Tom’s GuideReal people at AI companies may be reading your private chats — here's how to protect yourselfLiving near an AI data center might not hurt your home's value — here's what new research foundA few tech leaders are rebranding AI as ‘superintelligence’ — calling it that doesn’t make it accurateSign up to the Tom's AI Guide weekly newsletter summing up all the biggest AI news you need to know. Plus, analysis from our AI editors and tips on how to use the latest AI tools! Elton Jones covers AI for Tom’s Guide, and tests all the latest models, from ChatGPT to Gemini to Claude to see which tools perform best — and how they can improve everyday productivity.He is also an experienced tech writer who has covered video games, mobile devices, headsets, and now artificial intelligence for over a decade. Since 2011, his work has appeared in publications including The Christian Post, Complex, TechRadar, Heavy, and ONE37pm, with a focus on clear, practical analysis.Today, Elton focuses on making AI more accessible by breaking down complex topics into useful, easy-to-understand insights for a wide range of readers.