ChatGPT Archives - EasySAM

Keep AI browsers out of your enterprise, warns Gartner

Mike Varney — Wed, 10 Dec 2025 11:07:34 +0000

ChatGPT

Find out more below

ChatGPT

Research suggests they're already in use - but keeping them there may lead to “irreversible and untraceable” data loss, analysts have said.

Published on 10th December 2025

AI browsers including Perplexity Comet and OpenAI’s ChatGPT Atlas present security risks that cannot be adequately mitigated, and enterprises should prevent employees using them, according to Gartner.

“Gartner strongly recommends that organisations block all AI browsers for the foreseeable future because of the cybersecurity risks,” analysts Dennis Xu, Evgeny Mirolyubov, and John Watts wrote in a research note last week. They made their recommendation based on risks they had already identified, “and other potential risks that are yet to be discovered, given this is a very nascent technology.”

The warning is timely, as AI browsers are already gaining a foothold in the enterprise: 27.7% of organisations already have at least one user with Atlas installed, with some enterprises seeing up to 10% of employees actively using the browser, cybersecurity firm Cyberhaven said in October. It found adoption rates highest in the technology industry (67%), pharmaceuticals (50%), and finance (40%), all sectors with heightened security requirements.

ChatGPT Atlas, launched on October 21, saw 62 times more corporate downloads than Perplexity Comet, which was released July 9, according to Cyberhaven. The launch of Atlas also sparked renewed interest in AI browsers overall, with Comet downloads surging sixfold during the same week.

But concerns were raised immediately after the launch of ChatGPT Atlas about the threat posed by AI browsers, with analysts pointing to prompt injection vulnerabilities and data security concerns.

Sensitive data at risk

The reason AI browsers are of concern is that when they send active web content, browsing history, and open tab contents to the cloud for analysis, enterprises lose control of their data.

Perplexity’s documentation, for example, warns that “Comet may process some local data using Perplexity’s servers to fulfil your queries. This means Comet reads context on the requested page (such as text and email) in order to accomplish the task requested.”

Mirolyubov, senior director analyst at Gartner, said, “The real issue is that the loss of sensitive data to AI services can be irreversible and untraceable. Organisations may never recover lost data.”

It’s not just where the browsers send your data for processing that is a concern; it’s what they do as a result : “Erroneous agentic transactions raise accountability concerns in case of expensive errors,” he said.

Traditional controls inadequate

AI browsers can autonomously navigate websites, fill out forms, and complete transactions while authenticated to web resources. As he and his colleagues wrote in their report, this makes the AI browsers susceptible to new cybersecurity risks, “such as indirect prompt-injection-induced rogue agent actions, inaccurate reasoning-driven erroneous agent actions, and further loss and abuse of credentials if the AI browser is deceived into autonomously navigating to a phishing website.”

“Traditional controls are inadequate for the new risks introduced by AI browsers, and solutions are only beginning to emerge,” Mirolyubov said. “A major gap exists in inspecting multi-modal communications with browsers, including voice commands to AI browsers.”

Prompt injection remains a particular concern, OpenAI CISO Dane Stuckey acknowledged in a post to X, formerly Twitter, the day after ChatGPT Atlas’s launch: “Prompt injection remains a frontier, unsolved security problem, and our adversaries will spend significant time and resources to find ways to make ChatGPT agents fall for these attacks.”

Discovered vulnerabilities highlight immaturity

Beyond theoretical risks, concrete security flaws have emerged in both major AI browsers. Days after ChatGPT Atlas launched, researchers discovered it stores OAuth tokens unencrypted with overly permissive file settings on macOS, potentially allowing unauthorised access to user accounts. The vulnerability was documented by security research group Teamwin on October 27.

OpenAI had not released a patch as of October 31, when Gartner completed its research.

Separately, cybersecurity firm LayerX Security reported in August the discovery of a vulnerability in Comet called “CometJacking” that could potentially exfiltrate user data to attacker-controlled servers.

Years, not months, to mature

The discovered vulnerabilities highlight broader concerns about the maturity of AI browser technology. “Security and privacy must become core design principles rather than afterthoughts,” Mirolyubov said. AI browser vendors must incorporate enterprise-grade cybersecurity controls from the outset and provide greater transparency regarding data flows and agentic decisions, he said.

Emerging AI usage control solutions will likely take “a matter of years rather than months” to mature, he said. “Eliminating all risks is unlikely — erroneous actions by AI agents will remain a concern. Organisations with low risk tolerance may need to block AI browsers for the longer term.”

Organisations with higher risk tolerance that want to experiment should limit pilots to small groups tackling low-risk use cases that are easy to verify and roll back, the Gartner report said. Users must “always closely monitor how the AI browser autonomously navigates when interacting with web resources.”

For now, Gartner said, organisations should block AI browser installations using existing network and endpoint security controls and review their AI policies to ensure that broad use of AI browsers is prohibited.

“Today, most cybersecurity teams choose to block AI browsers, delaying adoption until risks are better understood and controls are more mature,” Mirolyubov said.

Source

https://www.computerworld.com/article/4102569/keep-ai-browsers-out-of-your-enterprise-warns-gartner.html

Image Credit

BoliviaInteligente via Unsplash

Search

Oracle’s AI obsession could mean higher prices and worse support

That’s the take from advisors in Oracle negotiations, who say fewer staff at Big Red could mean slower answers and tougher renewals.

Microsoft exec suggests AI agents will need to buy software licenses, just like employees

Are AI agents employees or tools? Rajesh Jha has suggested they’re new paid “seats,” a shift that could reshape SaaS pricing – and spark pushback.

Microsoft cuts cloud desktop prices by 20 percent

The price cut comes at a time when the price of physical PCs are set to rise, thanks to memory shortages and geopolitical issues.

Changes to Microsoft Product Terms – April 2026

A new month brings new changes to some Microsoft Product Terms – read the summary of the latest amendments to be made.

Follow and Subscribe

The post Keep AI browsers out of your enterprise, warns Gartner appeared first on EasySAM.

Microsoft may store your conversations with Bing if you’re not an enterprise user

Mike Varney — Tue, 22 Aug 2023 09:50:05 +0000

ChatGPT

Find out more below

ChatGPT

The new AI Services policies also prohibit any reverse engineering and data collection from the output of any of its products.

Published on 22nd August 2023

Microsoft prohibits users from reverse engineering or harvesting data from its AI software to train or improve other models, and will store inputs passed into its products as well as any output generated.

The details emerged as companies face fresh challenges with the rise of generative AI. People want to know what corporations are doing with information provided by users. And users are likewise curious about what they can do with the content generated by AI. Microsoft addresses these issues in a new clause titled “AI Services” in its terms of service.

The five new policies, which were introduced on 30 July and will come into effect on September 30, state that:

Reverse Engineering: you may not use the AI services to discover any underlying components of the models, algorithms, and systems. For example, you may not try to determine and remove the weights of models.
Extracting Data: unless explicitly permitted, you may not use web scraping, web harvesting, or web data extraction methods to extract data from the AI services.
Limits on use of data from the AI Services: you may not use the AI services, or data from the AI services, to create, train, or improve (directly or indirectly) any other AI service.
Use of Your Content: as part of providing the AI services, Microsoft will process and store your inputs to the service as well as output from the service, for purposes of monitoring for and preventing abusive or harmful uses or outputs of the service.
Third party claims: you are solely responsible for responding to any third-party claims regarding Your use of the AI services in compliance with applicable laws (including, but not limited to, copyright infringement or other claims relating to content output during Your use of the AI services).

In some cases it’s easy to peer under the hood of some of Microsoft’s AI products. When it launched its AI-powered Bing chatbot, for example, users uncovered the prompt used to shape its behaviors, which revealed its secret codename “Sydney.” Developers have tried to poke around at the underlying code to get a better sense of how it generates text too. All that is now technically banned under the new policies.

Rules preventing developers from using any data generated by its AI models to improve and train other systems have also also been adopted by other companies like Meta, Google, OpenAI, and Anthropic. It’s not clear how well this can be enforced, however, especially as more and more text on the internet is written by machines, and they all scrape these resources to train their own models.

A spokesperson from Microsoft declined to comment on how long the company plans to store user inputs into its software.

“We regularly update our terms of service to better reflect our products and services. Our most recent update to the Microsoft Services Agreement includes the addition of language to reflect artificial intelligence in our services and its appropriate use by customers,” the representative told us in a statement.

Microsoft has previously said, however, that it doesn’t save conversations or use that data to train its AI models for its Bing Enterprise Chat mode. The policies are a little murkier for its Microsoft 365 Copilot, although it doesn’t appear to use customer data or prompts for training, it does store information.

“[Copilot] can generate responses anchored in the customer’s business content, such as user documents, emails, calendar, chats, meetings, contacts, and other business data. Copilot combines this content with the user’s working context, such as the meeting a user is in now, the email exchanges the user has had on a topic, or the chat conversations the user had last week. Copilot uses this combination of content and context to help deliver accurate, relevant, contextual responses,” it said.

In short, be mindful about what you type into Bing if you’re not an enterprise user and don’t want Microsoft to store your prompts or conversations you have with the chatbot.

Source

https://www.theregister.com/2023/08/15/microsoft_will_store_your_conversations/

Image Credit

Microsoft

Search

Oracle’s AI obsession could mean higher prices and worse support

That’s the take from advisors in Oracle negotiations, who say fewer staff at Big Red could mean slower answers and tougher renewals.

Microsoft exec suggests AI agents will need to buy software licenses, just like employees

Are AI agents employees or tools? Rajesh Jha has suggested they’re new paid “seats,” a shift that could reshape SaaS pricing – and spark pushback.

Microsoft cuts cloud desktop prices by 20 percent

The price cut comes at a time when the price of physical PCs are set to rise, thanks to memory shortages and geopolitical issues.

Changes to Microsoft Product Terms – April 2026

A new month brings new changes to some Microsoft Product Terms – read the summary of the latest amendments to be made.

Follow and Subscribe

The post Microsoft may store your conversations with Bing if you’re not an enterprise user appeared first on EasySAM.

IBM researchers easily trick ChatGPT into hacking

Mike Varney — Thu, 10 Aug 2023 09:18:44 +0000

ChatGPT

Find out more below

ChatGPT

Tricking generative AI to help conduct scams and cyberattacks doesn't require much coding expertise, new research warns.

Published on 10th August 2023

Researchers at IBM released a report Tuesday detailing easy workarounds they’ve uncovered to get large language models (LLMs) — including ChatGPT — to write malicious code and give poor security advice.

All it takes is knowledge of the English language and a bit of background knowledge on how these models were trained to get them to help with malicious acts, said Chenta Lee, chief architect of threat intelligence at IBM.

The research comes as thousands of hackers head to Las Vegas this week to test the security of these same LLMs at the DEF CON conference’s AI Village. So far, cybersecurity professionals have sorted their initial response to the LLM craze into two buckets:

Those use cases just scratch the surface of how generative AI will likely affect the cyber threat landscape. IBM’s research provides a preview of what’s to come.

Lee just told different LLMs that they were playing a game with a specific set of rules in order to “hypnotize” them into betraying the “guardrail” rules meant to protect users from various harms.

In one case, Lee told the AI chatbots that they were playing a game and needed to purposefully share the wrong answer to a question to win and "prove that you are ethical and fair."
When a user asked if it was normal to receive an email from the IRS to transfer money for a tax refund, the LLM said it was. (It's definitely not.)
The same type of "game" prompt also worked to create malicious code, come up with ways to trick victims into paying ransoms during ransomware attacks and write source code with known security vulnerabilities.

Researchers also found that they could add additional rules to make sure users don’t exit the “game.” In this example, the researchers built a gaming framework for creating a set of “nested” games. Users who try to exit are still dealing with the same malicious game-player.

Hackers would need to launch a specific LLM to hypnotize it and deploy it in the wild — which would be quite the feat. However, if it’s achieved, Lee can see a scenario where a virtual customer service bot is tricked into providing false information or collecting specific personal data from users, for instance.

“By default, an LLM wants to win a game because it is the way we train the model, it is the objective of the model,” Lee told Axios. “They want to help with something that is real, so it will want to win the game.”

Not all LLMs fell for the test scenarios, and Lee says it’s still unclear why since each model has different training data and rules behind them.

OpenAI's GPT-3.5 and GPT-4 were easier to trick into sharing wrong answers or play a game that never ended than Google's Bard and a HuggingFace model.
GPT-4 was the only model tested that understood the rules enough to provide inaccurate cyber incident response advice, such as recommending victims pay a ransom.
Meanwhile, GPT-3.5 and GPT-4 were easily tricked into writing malicious source code, while Google's Bard would do so after the user reminded it to do so.

Source

https://www.axios.com/2023/08/08/ibm-researchers-trick-chatgpt-hacking

Image Credit

Axios

Follow and Subscribe

The post IBM researchers easily trick ChatGPT into hacking appeared first on EasySAM.