Search This Blog

Powered by Blogger.

Blog Archive

Labels

Footer About

Footer About

Labels

Showing posts with label ChatGPT. Show all posts

Some ChatGPT Browser Extensions Are Putting User Accounts at Risk

 


Cybersecurity researchers are cautioning users against installing certain browser extensions that claim to improve ChatGPT functionality, warning that some of these tools are being used to steal sensitive data and gain unauthorized access to user accounts.

These extensions, primarily found on the Chrome Web Store, present themselves as productivity boosters designed to help users work faster with AI tools. However, recent analysis suggests that a group of these extensions was intentionally created to exploit users rather than assist them.

Researchers identified at least 16 extensions that appear to be connected to a single coordinated operation. Although listed under different names, the extensions share nearly identical technical foundations, visual designs, publishing timelines, and backend infrastructure. This consistency indicates a deliberate campaign rather than isolated security oversights.

As AI-powered browser tools become more common, attackers are increasingly leveraging their popularity. Many malicious extensions imitate legitimate services by using professional branding and familiar descriptions to appear trustworthy. Because these tools are designed to interact deeply with web-based AI platforms, they often request extensive permissions, which exponentially increases the potential impact of abuse.

Unlike conventional malware, these extensions do not install harmful software on a user’s device. Instead, they take advantage of how browser-based authentication works. To operate as advertised, the extensions require access to active ChatGPT sessions and advanced browser privileges. Once installed, they inject hidden scripts into the ChatGPT website that quietly monitor network activity.

When a logged-in user interacts with ChatGPT, the platform sends background requests that include session tokens. These tokens serve as temporary proof that a user is authenticated. The malicious extensions intercept these requests, extract the tokens, and transmit them to external servers controlled by the attackers.

Possession of a valid session token allows attackers to impersonate users without needing passwords or multi-factor authentication. This can grant access to private chat histories and any external services connected to the account, potentially exposing sensitive personal or organizational information. Some extensions were also found to collect additional data, including usage patterns and internal access credentials generated by the extension itself.

Investigators also observed synchronized publishing behavior, shared update schedules, and common server infrastructure across the extensions, reinforcing concerns that they are part of a single, organized effort.

While the total number of installations remains relatively low, estimated at fewer than 1,000 downloads, security experts warn that early-stage campaigns can scale rapidly. As AI-related extensions continue to grow in popularity, similar threats are likely to emerge.

Experts advise users to carefully evaluate browser extensions before installation, pay close attention to permission requests, and remove tools that request broad access without clear justification. Staying cautious is increasingly important as browser-based attacks become more subtle and harder to detect.

AI Can Answer You, But Should You Trust It to Guide You?



Artificial intelligence tools are expanding faster than any digital product seen before, reaching hundreds of millions of users in a short period. Leading technology companies are investing heavily in making these systems sound approachable and emotionally responsive. The goal is not only efficiency, but trust. AI is increasingly positioned as something people can talk to, rely on, and feel understood by.

This strategy is working because users respond more positively to systems that feel conversational rather than technical. Developers have learned that people prefer AI that is carefully shaped for interaction over systems that are larger but less refined. To achieve this, companies rely on extensive human feedback to adjust how AI responds, prioritizing politeness, reassurance, and familiarity. As a result, many users now turn to AI for advice on careers, relationships, and business decisions, sometimes forming strong emotional attachments.

However, there is a fundamental limitation that is often overlooked. AI does not have personal experiences, beliefs, or independent judgment. It does not understand success, failure, or responsibility. Every response is generated by blending patterns from existing information. What feels like insight is often a safe and generalized summary of commonly repeated ideas.

This becomes a problem when people seek meaningful guidance. Individuals looking for direction usually want practical insight based on real outcomes. AI cannot provide that. It may offer comfort or validation, but it cannot draw from lived experience or take accountability for results. The reassurance feels real, while the limitations remain largely invisible.

In professional settings, this gap is especially clear. When asked about complex topics such as pricing or business strategy, AI typically suggests well-known concepts like research, analysis, or optimization. While technically sound, these suggestions rarely address the challenges that arise in specific situations. Professionals with real-world experience know which mistakes appear repeatedly, how people actually respond to change, and when established methods stop working. That depth cannot be replicated by generalized systems.

As AI becomes more accessible, some advisors and consultants are seeing clients rely on automated advice instead of expert guidance. This shift favors convenience over expertise. In response, some professionals are adapting by building AI tools trained on their own methods and frameworks. In these cases, AI supports ongoing engagement while allowing experts to focus on judgment, oversight, and complex decision-making.

Another overlooked issue is how information shared with generic AI systems is used. Personal concerns entered into such tools do not inform better guidance or future improvement by a human professional. Without accountability or follow-up, these interactions risk becoming repetitive rather than productive.

Artificial intelligence can assist with efficiency, organization, and idea generation. However, it cannot lead, mentor, or evaluate. It does not set standards or care about outcomes. Treating AI as a substitute for human expertise risks replacing growth with comfort. Its value lies in support, not authority, and its effectiveness depends on how responsibly it is used.

High Severity Flaw In Open WebUI Can Leak User Conversations and Data


A high-severity security bug impacting Open WebUI has been found by experts. It may expose users to account takeover (ATO) and, in some incidents, cause full server compromise. 

Talking about WebUI, Cato researchers said, “When a platform of this size becomes vulnerable, the impact isn’t just theoretical. It affects production environments managing research data, internal codebases, and regulated information.”

The flaw is tracked as CVE-2025-64496 and found by Cato Networks experts. The vulnerability affects Open WebUI versions 0.6.34 and older if the Director Connection feature is allowed. The flaw has a severity rating of 7.3 out of 10. 

The vulnerability exists inside Direct Connections, which allows users to connect Open WebUI to external OpenAI-supported model servers. While built for supporting flexibility and self-hosted AI workflows, the feature can be exploited if a user is tricked into linking with a malicious server pretending to be a genuine AI endpoint. 

Fundamentally, the vulnerability comes from a trust relapse between unsafe model servers and the user's browser session. A malicious server can send a tailored server-sent events message that prompts the deployment of JavaScript code in the browser. This lets a threat actor steal authentication tokens stored in local storage. When the hacker gets these tokens, it gives them full access to the user's Open WebUI account. Chats, API keys, uploaded documents, and other important data is exposed. 

Depending on user privileges, the consequences can be different.

Consequences?

  • Hackers can steal JSON web tokens and hijack sessions. 
  • Full account hack, this includes access to chat logs and uploaded documents.
  • Leak of important data and credentials shared in conversations. 
  • If the user has enabled workspace.tools permission, it can lead to remote code execution (RCE). 

Open WebUI maintainers were informed about the issue in October 2025, and publicly disclosed in November 2025, after patch validation and CVE assignment. Open WebUI variants 0.6.35 and later stop the compromised execute events, patching the user-facing threat.

Open WebUI’s security patch will work for v0.6.35 or “newer versions, which closes the user-facing Direct Connections vulnerability. However, organizations still need to strengthen authentication, sandbox extensibility and restrict access to specific resources,” according to Cato Networks researchers.





New US Proposal Allows Users to Sue AI Companies Over Unauthorised Data Use


US AI developers would be subject to data privacy obligations applicable in federal court under a wide legislative proposal disclosed recently by the US senate Marsha Blackburn, R-Tenn. 

About the proposal

Beside this, the proposal will create a federal right for users to sue companies for misusing their personal data for AI model training without proper consent. The proposal allows statutory and punitive damages, attorney fees and injunctions. 

Blackburn is planning to officially introduce the bill this year to codify President Donald Trump’s push for “one federal rule book” for AI, according to the press release. 

Why the need for AI regulations 

The legislative framework comes on the heels of Trump’s signing of an executive order aimed at blocking “onerous” AI laws at the state level and promoting a national policy framework for the technology.  

In order to ensure that there is a least burdensome national standard rather than fifty inconsistent State ones, the directive required the administration to collaborate with Congress. 

Michael Kratsios, the president's science and technology adviser, and David Sacks, the White House special adviser for AI and cryptocurrency, were instructed by the president to jointly propose federal AI legislation that would supersede any state laws that would contradict with administration policy. 

Blackburn stated in the Friday release that rather than advocating for AI amnesty, President Trump correctly urged Congress to enact federal standards and protections to address the patchwork of state laws that have impeded AI advancement.

Key highlights of proposal:

  • Mandate that regulations defining "minimum reasonable" AI protections be created by the Federal Trade Commission. 
  • Give the U.S. attorney general, state attorneys general, and private parties the authority to sue AI system creators for damages resulting from "unreasonably dangerous or defective product claims."
  • Mandate that sizable, state-of-the-art AI developers put procedures in place to control and reduce "catastrophic" risks associated with their systems and provide reports to the Department of Homeland Security on a regular basis. 
  • Hold platforms accountable for hosting an unauthorized digital replica of a person if they have actual knowledge that the replica was not authorized by the person portrayed.
  • Require quarterly reporting to the Department of Labor of AI-related job effects, such as job displacement and layoffs.

The proposal will preempt state laws regulating the management of catastrophic AI risks. The legislation will also mostly “preempt” state laws for digital replicas to make a national standard for AI. 

The proposal will not preempt “any generally applicable law, including a body of common law or a scheme of sectoral governance that may address” AI. The bill becomes effective 180 days after enforcement. 

Chinese Open AI Models Rival US Systems and Reshape Global Adoption

 

Chinese artificial intelligence models have rapidly narrowed the gap with leading US systems, reshaping the global AI landscape. Once considered followers, Chinese developers are now producing large language models that rival American counterparts in both performance and adoption. At the same time, China has taken a lead in model openness, a factor that is increasingly shaping how AI spreads worldwide. 

This shift coincides with a change in strategy among major US firms. OpenAI, which initially emphasized transparency, moved toward a more closed and proprietary approach from 2022 onward. As access to US-developed models became more restricted, Chinese companies and research institutions expanded the availability of open-weight alternatives. A recent report from Stanford University’s Human-Centered AI Institute argues that AI leadership today depends not only on proprietary breakthroughs but also on reach, adoption, and the global influence of open models. 

According to the report, Chinese models such as Alibaba’s Qwen family and systems from DeepSeek now perform at near state-of-the-art levels across major benchmarks. Researchers found these models to be statistically comparable to Anthropic’s Claude family and increasingly close to the most advanced offerings from OpenAI and Google. Independent indices, including LMArena and the Epoch Capabilities Index, show steady convergence rather than a clear performance divide between Chinese and US models. 

Adoption trends further highlight this shift. Chinese models now dominate downstream usage on platforms such as Hugging Face, where developers share and adapt AI systems. By September 2025, Chinese fine-tuned or derivative models accounted for more than 60 percent of new releases on the platform. During the same period, Alibaba’s Qwen surpassed Meta’s Llama family to become the most downloaded large language model ecosystem, indicating strong global uptake beyond research settings. 

This momentum is reinforced by a broader diffusion effect. As Meta reduces its role as a primary open-source AI provider and moves closer to a closed model, Chinese firms are filling the gap with freely available, high-performing systems. Stanford researchers note that developers in low- and middle-income countries are particularly likely to adopt Chinese models as an affordable alternative to building AI infrastructure from scratch. However, adoption is not limited to emerging markets, as US companies are also increasingly integrating Chinese open-weight models into products and workflows. 

Paradoxically, US export restrictions limiting China’s access to advanced chips may have accelerated this progress. Constrained hardware access forced Chinese labs to focus on efficiency, resulting in models that deliver competitive performance with fewer resources. Researchers argue that this discipline has translated into meaningful technological gains. 

Openness has played a critical role. While open-weight models do not disclose full training datasets, they offer significantly more flexibility than closed APIs. Chinese firms have begun releasing models under permissive licenses such as Apache 2.0 and MIT, allowing broad use and modification. Even companies that once favored proprietary approaches, including Baidu, have reversed course by releasing model weights. 

Despite these advances, risks remain. Open-weight access does not fully resolve concerns about state influence, and many users rely on hosted services where data may fall under Chinese jurisdiction. Safety is another concern, as some evaluations suggest Chinese models may be more susceptible to jailbreaking than US counterparts. 

Even with these caveats, the broader trend is clear. As performance converges and openness drives adoption, the dominance of US commercial AI providers is no longer assured. The Stanford report suggests China’s role in global AI will continue to expand, potentially reshaping access, governance, and reliance on artificial intelligence worldwide.

Adobe Brings Photo, Design, and PDF Editing Tools Directly Into ChatGPT

 



Adobe has expanded how users can edit images, create designs, and manage documents by integrating select features of its creative software directly into ChatGPT. This update allows users to make visual and document changes simply by describing what they want, without switching between different applications.

With the new integration, tools from Adobe Photoshop, Adobe Acrobat, and Adobe Express are now available inside the ChatGPT interface. Users can upload images or documents and activate an Adobe app by mentioning it in their request. Once enabled, the tool continues to work throughout the conversation, allowing multiple edits without repeatedly selecting the app.

For image editing, the Photoshop integration supports focused and practical adjustments rather than full professional workflows. Users can modify specific areas of an image, apply visual effects, or change settings such as brightness, contrast, and exposure. In some cases, ChatGPT presents multiple edited versions for users to choose from. In others, it provides interactive controls, such as sliders, to fine-tune the result manually.

The Acrobat integration is designed to simplify common document tasks. Users can edit existing PDF files, reduce file size, merge several documents into one, convert files into PDF format, and extract content such as text or tables. These functions are handled directly within ChatGPT once a file is uploaded and instructions are given.

Adobe Express focuses on design creation and quick visual content. Through ChatGPT, users can generate and edit materials like posters, invitations, and social media graphics. Every element of a design, including text, images, colors, and animations, can be adjusted through conversational prompts. If users later require more detailed control, their projects can be opened in Adobe’s standalone applications to continue editing.

The integrations are available worldwide on desktop, web, and iOS platforms. On Android, Adobe Express is already supported, while Photoshop and Acrobat compatibility is expected to be added in the future. These tools are free to use within ChatGPT, although advanced features in Adobe’s native software may still require paid plans.

This launch follows OpenAI’s broader effort to introduce third-party app integrations within ChatGPT. While some earlier app promotions raised concerns about advertising-like behavior, Adobe’s tools are positioned as functional extensions rather than marketing prompts.

By embedding creative and document tools into a conversational interface, Adobe aims to make design and editing more accessible to users who may lack technical expertise. The move also reflects growing competition in the AI space, where companies are racing to combine artificial intelligence with practical, real-world tools.

Overall, the integration represents a shift toward more interactive and simplified creative workflows, allowing users to complete everyday editing tasks efficiently while keeping professional software available for advanced needs.




OpenAI Warns Future AI Models Could Increase Cybersecurity Risks and Defenses

 

Meanwhile, OpenAI told the press that large language models will get to a level where future generations of these could pose a serious risk to cybersecurity. The company in its blog postingly admitted that powerful AI systems could eventually be used to craft sophisticated cyberattacks, such as developing previously unknown software vulnerabilities or aiding stealthy cyber-espionage operations against well-defended targets. Although this is still theoretical, OpenAI has underlined that the pace with which AI cyber-capability improvements are taking place demands proactive preparation. 

The same advances that could make future models attractive for malicious use, according to the company, also offer significant opportunities to strengthen cyber defense. OpenAI said such progress in reasoning, code analysis, and automation has the potential to significantly enhance security teams' ability to identify weaknesses in systems better, audit complex software systems, and remediate vulnerabilities more effectively. Instead of framing the issue as a threat alone, the company cast the issue as a dual-use challenge-one in which adequate management through safeguards and responsible deployment would be required. 

In the development of such advanced AI systems, OpenAI says it is investing heavily in defensive cybersecurity applications. This includes helping models improve particularly on tasks related to secure code review, vulnerability discovery, and patch validation. It also mentioned its effort on creating tooling supporting defenders in running critical workflows at scale, notably in environments where manual processes are slow or resource-intensive. 

OpenAI identified several technical strategies that it thinks are critical to the mitigation of cyber risk associated with increased capabilities of AI systems: stronger access controls to restrict who has access to sensitive features, hardened infrastructure to prevent abuse, outbound data controls to reduce the risk of information leakage, and continuous monitoring to detect anomalous behavior. These altogether are aimed at reducing the likelihood that advanced capabilities could be leveraged for harmful purposes. 

It also announced the forthcoming launch of a new program offering tiered access to additional cybersecurity-related AI capabilities. This is intended to ensure that researchers, enterprises, and security professionals working on legitimate defensive use cases have access to more advanced tooling while providing appropriate restrictions on higher-risk functionality. Specific timelines were not discussed by OpenAI, although it promised that more would be forthcoming very soon. 

Meanwhile, OpenAI also announced that it would create a Frontier Risk Council comprising renowned cybersecurity experts and industry practitioners. Its initial mandate will lie in assessing the cyber-related risks that come with frontier AI models. But this is expected to expand beyond this in the near future. Its members will be required to offer advice on the question of where the line should fall between developing capability responsibly and possible misuse. And its input would keep informing future safeguards and evaluation frameworks. 

OpenAI also emphasized that the risks of AI-enabled cyber misuse have no single-company or single-platform constraint. Any sophisticated model, across the industry, it said, may be misused if there are no proper controls. To that effect, OpenAI said it continues to collaborate with peers through initiatives such as the Frontier Model Forum, sharing threat modeling insights and best practices. 

By recognizing how AI capabilities could be weaponized and where the points of intervention may lie, the company believes, the industry will go a long way toward balancing innovation and security as AI systems continue to evolve.

IDESaster Report: Severe AI Bugs Found in AI Agents Can Lead to Data Theft and Exploit


Using AI agents for data exfiltrating and RCE

A six-month research into AI-based development tools has disclosed over thirty security bugs that allow remote code execution (RCE) and data exfiltration. The findings by IDEsaster research revealed how AI agents deployed in IDEs like Visual Studio Code, Zed, JetBrains products and various commercial assistants can be tricked into leaking sensitive data or launching hacker-controlled code. 

The research reports that 100% of tested AI IDEs and coding agents were vulnerable. Impacted products include GitHub, Windsurf, Copilot, Cursor, Kiro.dev, Zed.dev, Roo Code, Junie, Cline, Gemini CLI, and Claude Code. At least twenty-four assigned CVEs and additional AWS advisories were also included. 

AI assistants exploitation 

The main problem comes from the way AI agents interact with IDE features. Autonomous components that could read, edit, and create files were never intended for these editors. Once-harmless features turned become attack surfaces when AI agents acquired these skills. In their threat model, all AI IDEs essentially disregard the base software. Since these features have been around for years, they consider them to be naturally safe. 

Attack tactic 

However, the same functionalities can be weaponized into RCE primitives and data exfiltration once autonomous AI bots are included. The research reported that this is an IDE-agnostic attack chain. 

It begins with context hacking via prompt-injection. Covert instructions can be deployed in file names, rule files, READMEs, and outputs from malicious MCP servers. When an agent reads the context, the tool can be redirected to run authorized actions that activate malicious behaviours in the core IDE. The last stage exploits built-in features to steal data or run hacker code in AI IDEs sharing core software layers.

Examples

Writing a JSON file that references a remote schema is one example. Sensitive information gathered earlier in the chain is among the parameters inserted by the agent that are leaked when the IDE automatically retrieves that schema. This behavior was seen in Zed, JetBrains IDEs, and Visual Studio Code. The outbound request was not suppressed by developer safeguards like diff previews.  

Another case study uses altered IDE settings to show complete remote code execution. An attacker can make the IDE execute arbitrary code as soon as a relevant file type is opened or created by updating an executable file that is already in the workspace and then changing configuration fields like php.validate.executablePath. Similar exposure is demonstrated by JetBrains utilities via workspace metadata.

According to the IDEsaster report, “It’s impossible to entirely prevent this vulnerability class short-term, as IDEs were not initially built following the Secure for AI principle. However, these measures can be taken to reduce risk from both a user perspective and a maintainer perspective.”