OpenAI CEO Sam Altman says the world is still far away from artificial general intelligence (AGI), despite the release of GPT-5, the company’s most advanced AI model yet.

According to Altman, while GPT-5 marks a significant leap in capability and general intelligence, it still falls short of what could be considered true AGI — a system that can match or exceed human reasoning across the board.

Altman, who has been one of the most vocal proponents of AGI development, stressed that OpenAI’s core mission to build artificial general intelligence that benefits all of humanity remains unchanged. However, he acknowledged that GPT-5, despite its power and versatility, still lacks several key features that he believes are essential to AGI.

Among those missing pieces is the ability to learn continuously from its environment. “This is not a model that learns as it’s deployed,” Altman noted, calling that absence one of the major indicators that OpenAI has not yet reached AGI.

He emphasized that the system does not improve itself dynamically by drawing from real-world interactions after training — something he sees as critical to what would qualify as general intelligence.

To understand the weight of Altman’s comments, it’s worth looking at how far the company has come. The first iteration of GPT, short for generative pre-trained transformer, was released in 2018, offering modest capabilities in text generation. Then came GPT-2 and GPT-3, with exponential improvements, culminating in GPT-4, which stunned the world with its ability to write, reason, code, and hold nuanced conversations. GPT-5 builds on all of this, introducing multimodal abilities, broader context windows, and a significantly more fluid understanding of human language and intention. Yet, even with all of that, Altman says there’s something missing — or many things.

The comments come at a time when debates around AGI are becoming more intense, with some voices warning that the tech industry is rushing blindly into a future it doesn’t fully understand, while others remain skeptical that AGI is anywhere close. Altman’s assessment threads the needle between optimism and caution. He says GPT-5 is a major step forward, “a significant fraction of the way to something very AGI-like,” as he put it, but not the finish line.

He also pointed to what comes next: superintelligence. For Altman, AGI is not the ceiling — it’s a milestone. His broader vision includes the creation of tools that not only think like humans but vastly outthink them, systems that could accelerate scientific breakthroughs and reshape the future of civilization.

What’s Under the Hood of GPT-5

GPT5 is a unified system with multiple internal models — a fast responder, a deeper reasoning variant, and a high-capacity “thinking-pro” edition. A real-time router dynamically selects the best model for each query, effectively replacing the need for users to manually pick the right tool for the task. Variant tiers — regular, mini, and nano — cater to different usage needs and budget levels.

In ChatGPT, GPT-5 is accessible to all users:

• Free tier: Standard GPT-5 with usage caps and automatic fallback to GPT-5 mini when limits are hit.
• Plus users: Higher usage thresholds and selectable access to “GPT-5 thinking.”
• Pro tier ($200/month): Unlimited access, including the more powerful “GPT-5 Pro.”

Developers using OpenAI’s API can choose among models scaled for cost, from full GPT-5 to budget-friendly nano options, with input/output token pricing that varies accordingly.

What’s Improved

OpenAI claims GPT-5 delivers state-of-the-art performance across domains:

• In coding, it excels at building responsive UIs, debugging large codebases, and executing long, agentic tasks.
• In writing, it handles structural ambiguity and literary nuance with greater fidelity.
• In health, it acts more like an active thought partner, probing context, flagging concerns, and scoring well on benchmarks like HealthBench.
• It also achieves top marks in math, science, vision, and multimodal reasoning.

Industry critics believe GPT-5 represents a pivotal shift in usability — a model that now truly feels like interacting with a PhD-level expert across domains. It also sets a new standard for practical, versatile AI assistance. Yet as OpenAI emphasizes, ambition remains on superintelligence — a theoretical level of AI exceeding human capacity and transformative potential in science, innovation, and society.

Nvidia’s Chief Security Officer, David Reber Jr., has issued a strong public denial of claims suggesting the company’s chips contain secret backdoors or kill switches, saying, “there is no such thing as a ‘good’ secret backdoor—only dangerous vulnerabilities that need to be eliminated.”

His remarks come as the company finds itself under growing pressure from both sides of the geopolitical divide, with US lawmakers advocating tighter control over chip exports and Chinese officials accusing the chipmaker of embedding covert access mechanisms in its AI hardware.

“To mitigate the risk of misuse, some pundits and policymakers propose requiring hardware ‘kill switches’ or built-in controls that can remotely disable GPUs without user knowledge and consent,” wrote Reber Jr. “Some suspect they might already exist,” he continues, in a nod to a probe already launched in China over alleged “loopholes and backdoor” vulnerabilities in the H20 chips that have been sold in the country.

The controversy stems from a convergence of domestic political pressure in the United States and intensifying scrutiny from China. In May, a bipartisan group of US lawmakers introduced the Chip Security Act, legislation that would require AI chipmakers like Nvidia to include embedded tracking systems capable of identifying unauthorized exports and potentially disabling hardware remotely. Though pitched as a national security safeguard, the bill leaves open the possibility of adding kill switches—components that could render the chips inoperable from afar.

While the bill is still in early legislative stages, it has already sparked concern among global buyers, especially in China, where regulators have accused Nvidia of secretly inserting such mechanisms into its AI chips. The allegations culminated in a public summons by Beijing’s cybersecurity watchdog, which demanded an explanation over what it described as “loopholes and backdoor” vulnerabilities in Nvidia’s H20 chips—products that were specially redesigned for China after Washington banned the sale of its most powerful AI processors, such as the A100 and H100.

The H20 chips were developed as a workaround to the US export restrictions imposed in 2022 and further tightened in 2023, which aimed to slow China’s progress in artificial intelligence and supercomputing. After months of review, the US government approved the H20 for limited export to China, signaling a green light for Nvidia to re-enter a lucrative market that had once been its largest revenue generator outside the US.

However, Beijing’s approval came with strings attached. Upon allowing the import of the modified chips, Chinese regulators initiated a technical review to ensure the chips did not contain any features that would allow them to be tracked, controlled, or shut down remotely. Officials in Beijing have warned that such components would constitute a national security threat and violate China’s sovereignty.

The probe is ongoing, and analysts say it underscores China’s deepening mistrust of US technology providers in the wake of broader Washington sanctions and scrutiny against firms like Huawei and TikTok.

Reber Jr.’s blog post appears to be a strategic defense against growing doubts from both Washington and Beijing. Addressing US policymakers directly, he wrote that mandatory kill switches would be “an open invitation for disaster” and “not sound policy.” His comments suggest that Nvidia is caught in a precarious position, attempting to reassure lawmakers in Washington of its compliance while also trying to calm fears in Beijing that it could be used as an extension of US intelligence interests.

AI chip sales are driving Nvidia’s explosive growth, and its ability to operate in China, still the world’s second-largest economy, could determine whether it remains dominant in the long run. But that dominance is now under threat. China is fast-tracking its own semiconductor ecosystem, with firms like Huawei rapidly advancing domestic alternatives to Nvidia’s technology. Analysts say that if Chinese regulators conclude that Nvidia’s chips are compromised or politically risky, it could accelerate Beijing’s decoupling from US tech suppliers.

Ironically, Huawei’s own fall from grace began under similar suspicions—that it had built backdoors into its telecom equipment for Chinese government access. Now the tables are turning, with Chinese regulators warning that it is Nvidia who may be giving foreign governments covert entry points into critical computing infrastructure.

Both Washington and Beijing appear determined to bend Nvidia’s business to their respective strategic goals. The US wants guarantees that its chips won’t fall into the hands of adversaries, while China wants assurance that US tech won’t be used as a geopolitical Trojan horse. Caught in the middle, Nvidia is trying to convince both sides that its hardware is neutral and secure—a hard sell in today’s fractured tech landscape.

OpenAI’s much-anticipated GPT-5 model officially launched Thursday, and Microsoft is rolling it out across nearly all of its major platforms in a sweeping upgrade that underscores the deepening AI collaboration between the two tech giants.

From Copilot in Windows to Microsoft 365, GitHub, and Azure AI Foundry, GPT-5 is now embedded across Microsoft’s ecosystem, making the latest generation of OpenAI’s large language model accessible to both free and paid users, depending on the product.

GPT-5 Comes to Microsoft Copilot — and It’s Free

One of the most immediate upgrades is in Microsoft Copilot, where GPT-5 is now the backbone of a new “smart mode”. This feature allows the AI assistant to intelligently switch between models depending on the task — for instance, choosing a more advanced model like GPT-5 for tasks that require deeper reasoning or problem-solving, and a lighter one for quicker responses.

Free users of Copilot will now get access to GPT-5 by default, mirroring OpenAI’s decision to offer GPT-5 access to free ChatGPT users. This means millions of Windows users interacting with Copilot in apps like Notepad, Paint, and File Explorer will now have the power of GPT-5 at their fingertips.

Microsoft 365 Copilot Gets Smarter

Enterprise users aren’t left out. Microsoft 365 Copilot, which integrates AI into apps like Word, Excel, Outlook, and Teams, is also getting GPT-5 today. Microsoft says the upgrade brings improved reasoning, better context retention, and more accurate responses during long conversations.

“With GPT-5, Microsoft 365 Copilot is better at reasoning through complex questions, staying on track in longer conversations and understanding the user’s context,” Microsoft said in a blog post.

Copilot Studio, Microsoft’s platform for customizing AI workflows and building plugins, also now runs GPT-5.

GitHub Copilot Upgrades to GPT-5

Developers are also getting a major boost. GitHub Copilot, Microsoft’s AI coding assistant, is rolling out GPT-5 to all paid plans starting today. The model delivers notable improvements in code generation, refactoring, and multi-step logic tasks, which GitHub says should result in fewer hallucinations and more production-grade output.

OpenAI confirmed that GPT-5 comes in four variants, with enhancements in both reasoning and code quality. One of those variants, GPT-5-chat, is optimized specifically for enterprise applications, including multimodal and context-aware conversations — a perfect fit for integrated dev environments and productivity tools.

Azure AI Foundry Opens GPT-5 to Developers

Rounding out the release, Microsoft is also making GPT-5 available on Azure AI Foundry — its developer platform for building and scaling AI-powered apps.

Using the new model router feature, developers can now automatically direct tasks to the best-fitting GPT-5 variant, ensuring efficiency for workloads that vary in complexity and modality. This integration gives startups and enterprises alike access to the most advanced AI available, without needing to build infrastructure from scratch.

A Unified AI Front

Today’s rollout of GPT-5 marks a milestone in Microsoft’s multi-year AI strategy. By embedding OpenAI’s most powerful model across its platforms, from personal productivity to enterprise development, Microsoft is building what it calls a “unified AI fabric”. That strategy puts the company in a strong position to maintain its lead in AI infrastructure, while pushing broader adoption of generative tools in both consumer and commercial markets.

With GPT-5 now live in both Microsoft and OpenAI products, a new phase of AI-powered computing has officially begun.

President Donald Trump on Thursday publicly called for the resignation of Intel CEO Lip-Bu Tan, intensifying scrutiny over the executive’s alleged ties to Chinese firms and throwing one of America’s most critical tech companies into crisis.

“The CEO of INTEL is highly CONFLICTED and must resign, immediately. There is no other solution to this problem,” Trump posted on Truth Social around 7:30 a.m., in a direct attack that jolted markets and caught Intel off guard. The company’s shares dropped 3.1% by the end of trading.

The controversy erupted after a Republican senator raised concerns about Tan’s reported financial interests in Chinese companies. Trump’s reaction, however, elevated the matter from a routine vetting issue to a full-blown political and public relations storm — one that now threatens to undermine confidence in Intel at a time when the U.S. is fighting to regain its global chipmaking edge.

In a statement sent to reporters hours later, an Intel spokesperson said: “Intel, the Board of Directors, and Lip-Bu Tan are deeply committed to advancing US national and economic security interests and are making significant investments aligned with the President’s America First agenda.”

But for many observers, the damage may already be underway. Veteran crisis-communications experts warn that the situation could spiral quickly unless Intel takes decisive steps to control the narrative.

While Trump has long targeted government officials, corporate lawyers, and university presidents, it is rare for the President to directly call for the ouster of a sitting CEO of a major American corporation. In this case, the attack comes amid the administration’s broader crackdown on perceived Chinese influence in high-tech sectors and as it pushes for the return of advanced manufacturing jobs to the U.S.

“President Trump remains fully committed to safeguarding our country’s national and economic security,” said a White House official in a statement to Business Insider. “This includes ensuring that iconic American companies in cutting-edge sectors are led by men and women who Americans can trust.”

Lip-Bu Tan, a veteran of the semiconductor industry, was appointed as Intel’s chief executive in March. Though seen by some as a strategic and globally connected leader, his past investment ties have now drawn fire at the highest levels.

Crisis Management Experts Weigh In

Crisis management experts who spoke to Business Insider said Intel needs to address the issue head-on to avert further crises. The U.S. chipmaker has tried to overcome its years of downturn, marked by relentless revenue declines. Tan was appointed the company’s CEO in a bid to revive its fortune.

Kevin Donahue, a 30-year veteran of corporate crisis communication, warned that Intel must act immediately.

“It’s potentially fatal,” Donahue said. “You absolutely cannot stay mum in situations like this.”

He urged Intel to issue a robust and multi-front response addressing investors, customers, and government stakeholders, while anticipating how the narrative might evolve online. “Otherwise,” he said, “the crisis grows.”

Another industry expert, Michele Ehrhart, formerly VP of global communications at FedEx and author of Crisis Compass, said Intel must decide whether to publicly back Tan or not.

“You have to really draw a line in the sand on what it is you’re trying to accomplish and how you’re planning to do that,” said Ehrhart, who is the author of the book “Crisis Compass.”

Ehrhart noted that putting Tan in front of the press could help restore trust, though she cautioned against over-communicating.

“Here’s our statement. We’re not going to go beyond it,” she advised.

Donahue also recommended that Intel assemble a rapid-response team of legal, PR, and government affairs experts to coordinate messaging and demonstrate the board’s confidence in Tan’s appointment.

“That’s a tall order,” he admitted.

Managing Trump’s Attack

But Trump’s involvement adds a uniquely unpredictable element. Evan Nierman, CEO of crisis PR firm Red Banyan, said dealing with a standard political figure is one thing, but confronting Trump is another matter entirely.

“You have to create an offramp,” Nierman said, advising Intel to find a diplomatic way to de-escalate. “Get in touch with people close to the President and try to demonstrate alignment.”

Nierman called Trump’s political style “transactional” and suggested that a careful review of his posts and speeches might reveal what Intel could say or offer to get him to step back. “You’re going to have a lot more success than telling him no, ignoring him, or trying to defend yourself.”

A New Test for Corporate America

The episode underscores a new reality for corporate America: doing business, especially in sensitive sectors like semiconductors, now involves navigating high-stakes geopolitical tensions, domestic political pressures, and the personal attention of the President himself.

For Intel, how it handles this moment could determine not just the future of its leadership but also its credibility with investors, regulators, and allies at a time when the U.S. is making historic investments in reshoring chip production.

As of Thursday evening, Intel’s board had not announced any change in leadership. But few expect this to fade quietly, not while the President continues to post.