Author: Andrew Hoblitzell

Earlier in the week, OpenAI announced it is restructuring its AI roadmap, consolidating its efforts around GPT-5 while scrapping the previously planned o3 standalone release.

Initially, OpenAI had intended to roll out GPT-4.5 (code-named Orion) as an intermediary update, but internal reports suggested that its improvements over GPT-4 were marginal. Instead, CEO Sam Altman confirmed that OpenAI will deliver a “unified intelligence” model with GPT-5, eliminating the need for users to select different versions within ChatGPT.

Altman has said GPT-5 will be designed to integrate multiple capabilities—reasoning, voice synthesis, search, and deep research—into a single model. OpenAI aims to streamline its product lineup by removing the current model picker and letting the AI dynamically determine how much computational power is needed for a given task. This move follows OpenAI’s increasing focus on reasoning models, which self-verify outputs for greater reliability at the cost of additional inference time. While OpenAI pioneered this approach with o1, competitors like DeepSeek have quickly closed the gap, prompting OpenAI to accelerate its release schedule to maintain its lead.

“We want to do a better job of sharing our intended roadmap, and a much better job simplifying our product offerings… We want AI to just work for you; we realize how complicated our model and product offerings have gotten.” – Sam Altman

Altman outlined OpenAI’s planned subscription tiers for GPT-5, where free users will have access to a standard intelligence level, while ChatGPT Plus and Pro subscribers will receive progressively more advanced reasoning capabilities. This marks a shift in OpenAI’s monetization strategy, moving beyond usage limits toward tiered model intelligence.

Technical hurdles remain, particularly around inference efficiency. As reasoning models require additional processing to validate outputs, computational costs rise, and latency becomes a concern. OpenAI, already reliant on Microsoft Azure for cloud resources, must balance these factors while scaling GPT-5’s capabilities. Additionally, growing competition from open-source AI models threatens OpenAI’s position in the market, as organizations increasingly seek customizable, locally hosted AI solutions. Indeed, while Altman has discussed GPT 4.5 getting good reception from testers, AI developer Elvis Saravia has also playfully noted tomorrow Altman may also comment GPT 4.5 ‘still has a long way to go’.

This comes at a time when Anthropic has also been developing a hybrid AI model that dynamically adjusts reasoning power, allowing developers to control computational resources via a sliding scale, unlike OpenAI’s fixed low-medium-high settings. This model excels in business-focused coding tasks, particularly in handling large codebases and generating accurate code, outperforming OpenAI’s top reasoning model in certain programming benchmarks. Some have speculated OpenAI, recognizing Anthropic’s approach, plans to merge its reasoning and traditional models into a single AI, potentially following Anthropic’s lead. Anthropic reportedly plans to release its new model in the coming weeks.

The competition for AI talent has been fierce, with OpenAI and its rivals aggressively trying to retain and attract top researchers. Thrive Capital, a major OpenAI investor, recently presented to OpenAI staff, emphasizing the potential financial losses of leaving for a startup, given OpenAI’s rapid valuation growth. This comes amid high-profile departures, including ex-research head Ilya Sutskever who left for Safe Superintelligence (SSI) and ex-CTO Mira Murati, who has recruited multiple OpenAI researchers for her new venture, Thinking Machines Lab. 

After it was reported that OpenAI has started rolling out its new GPT Store, it was also discovered that some of the data they’re built on is easily exposed. Multiple groups have begun finding that the system has the potential to leak otherwise sensitive information.

Researchers from Northwestern University wrote “through comprehensive testing of over 200 user-designed GPT models via adversarial prompts, we demonstrate that these systems are susceptible to prompt injections.” Prompt injection attacks are a known vulnerability vector to extract sensitive information or manipulate the model’s output. Prompt injection is just one at vulnerability for language models, with some others including prompt leaking and jailbreaking.

Prompt injection refers to the technique where an attacker crafts specific inputs or ‘prompts’ to manipulate the behavior of Large Language Models (LLMs) like GPTs. In their research, the team found that through prompt injection, “an adversary can not only extract the customized system prompts but also access uploaded files.”

Yesterday I created a custom GPT for http://Levels.fyi with a limited subset of our data as a knowledge source (RAG). The feedback was incredible and folks got super creative with some of their prompts. However, I found out shortly that the source data file was leaking. – Zuhayeer Musa

“Our findings underscore the urgent need for robust security frameworks in the design and deployment of customizable GPT models.” They emphasize that while customization opens up new frontiers in AI utility by allowing individuals and organizations to create AI models tailored for specific needs without necessitating coding skills; it also introduces additional vectors for potential security vulnerabilities.

“We identified key security risks related to prompt injection and conducted an extensive evaluation… Our tests revealed that these prompts could almost entirely expose system prompts and retrieve uploaded files from most custom GPTS.” This indicates a significant vulnerability in current custom GPTs regarding system prompt extraction and file disclosure.

The researchers conclude their paper with an urgent call to action: “Our findings highlight the urgent need for enhanced security measures in the rapidly evolving domain of customizable AI, and we hope this sparks further discussion on the subject.” They stress that as AI technologies continue to evolve, it is crucial to strike a balance between innovation and security.

Adversa AI also recently showed that GPTs may be tricked into leaking how they were built, including prompts, API names, and uploaded document metadata and content. OpenAI noted it had patched the vulnerabilities reported by the Adversa AI researchers.

“We’re constantly working to make our models and products safer and more robust against adversarial attacks, including prompt injections, while also maintaining the models’ usefulness and task performance.” – OpenAI to Wired

Developers interested in learning more about the work from Northwestern University may read the technical paper on Arxiv or follow reaction on social media.