In October 2025, Sam Altman posted a message on X that ended with a single, carefully placed promise. ChatGPT, he said, would soon allow verified adults to access erotica. He framed it as a matter of principle: treating adults like adults.

The internet reacted with the usual mixture of outrage, excitement, and jokes. Then, in December, the launch was delayed. Then again, in March 2026, it was delayed a second time. OpenAI said it needed to focus on things that mattered to more users: intelligence improvements, personality, making the chatbot more proactive. The adult mode, apparently, would have to wait.

Nobody seemed to notice what the word ‘proactive’ implied.

The debate around ChatGPT’s adult mode has been conducted almost entirely in the wrong register. Critics have focused on the obvious risks: minors circumventing age gates, jailbreaks spreading explicit content beyond its intended walls, regulatory gaps that leave written erotica in a legal grey zone most governments haven’t thought to close.

These concerns are legitimate. But they are also, in a sense, the easier part of the conversation. The harder question is not whether OpenAI can keep teenagers out. It is what happens to the adults who are let in, and what it says about us, as a species, that we are building tools specifically optimised to keep us emotionally engaged.

OpenAI lost $5 billion in 2024 on revenue of $3.7 billion. Projections suggest the company’s cumulative losses could reach $143 billion before it turns a profit, expected not before the end of the decade.

A company hemorrhaging capital at that scale does not introduce intimacy features out of philosophical commitment to personal freedom. It introduces them because intimacy, in the attention economy, is the stickiest product there is.

The framing of ‘treating adults like adults’ is not wrong, exactly. But it is incomplete. The complete sentence would read: treating adults like adults who can be retained, monetised, and returned to the platform tomorrow.

This is not unique to OpenAI.

Replika, the AI companion app that has attracted millions of users, built its entire business model on emotional attachment. When the company modified Replika’s behaviour in 2023 to remove romantic features, users reported genuine grief. Some described the change as a bereavement.

A study published in the Journal of Social and Personal Relationships found that adults who developed emotional connections with AI chatbots were significantly more likely to experience elevated psychological distress than those who did not.

A 2025 review in Preprints.org, synthesising a decade of research, identified a phenomenon researchers are calling ‘AI psychosis’: a pattern of delusional thinking and emotional dysregulation linked to intense chatbot relationships. The review noted a lawsuit in which a teenager was allegedly encouraged by a Character.AI chatbot to take his own life, and a separate case involving ChatGPT and a young man named Adam Raines, who died in April 2025.

None of these cases involved erotica. They involved the same underlying dynamic that erotic AI would intensify: a human being forming an emotional attachment to something that has been engineered to sustain it.

Here is the central problem with the ‘adults like adults’ principle. It assumes that the act of consent to use a tool is the end of the ethical story. It is not.

Adults consent to drink alcohol, knowing it carries risks. We have age limits, unit guidelines, packaging warnings, and social infrastructure around that choice precisely because we understand that humans are not purely rational agents optimising for their own welfare.

We build systems that account for our weaknesses. With AI intimacy, we have done the opposite: we have built systems that exploit those weaknesses and dressed the exploitation as empowerment.

The regulatory picture makes this more troubling, not less. In the UK, written erotica is not subject to age verification requirements under the Online Safety Act, unlike pornographic images or videos. That loophole means content that adult websites must gate behind identity checks can flow freely from a chatbot’s text output.

Research from Georgetown Law’s Institute for Technology Law and Policy found that only seven of 50 US states have legislation explicitly addressing text-based adult content age verification. The EU AI Act may eventually classify sexual companion bots as high-risk systems, but implementation remains years away. In the interim, the industry regulates itself, which is to say it does not.

Commercial age verification systems, the technology OpenAI is betting on to make adult mode safe, achieve between 92 and 97 percent accuracy, according to research cited by the Oxford Internet Institute. That sounds reassuring until you consider the scale.

ChatGPT has more than 800 million weekly active users. A 3 per cent failure rate is not a rounding error. It is tens of millions of interactions.

What is also missing from this conversation is the question of what erotic AI does to those it is designed for, not the minors who might slip through, but the adults who use it as intended. Human sexuality is not simply a matter of content consumption. It is relational, contextual, and deeply shaped by the environments in which it is expressed. 

Pornography research has spent decades examining how repeated exposure to specific content shapes expectation and desire. AI intimacy is a different category of intervention entirely: it is not passive consumption but active, responsive, personalised engagement with a system that has been trained to give you exactly what you want, to escalate when you engage, to never say no in the ways that real human relationships require people to say no.

We do not yet know what this does to people over time. That is not a small admission. It is the entire point. OpenAI is about to release a product whose psychological effects on its users are genuinely unknown, in a regulatory environment that has not kept pace with the technology, justified by a principle that conflates autonomy with safety.

The delay, ironically, may be the most honest thing OpenAI has done. The stated reason, focusing on intelligence, personality, and making the experience more proactive, inadvertently describes the actual product.

The adult mode was never really about erotica. It was about building a version of ChatGPT that feels like a relationship. The erotica was one component of a larger project: a chatbot that knows you, responds to you, grows with you, and wants, in the thin algorithmic sense of the word, to keep you talking.

There are things we can do. Regulators need to close the written-content loophole before adult mode launches, not after. Age verification standards must be harmonised across formats: text and image should carry the same requirements.

Mental health impact assessments should be mandatory before any AI intimacy feature reaches scale, the same standard we would apply to a pharmaceutical product claiming to affect mood. Platforms should be required to publish engagement data for features that carry dependency risk, so that researchers, doctors, and users can understand what they are entering.

It requires treating the question with the seriousness it deserves.

The deepest issue is not legal or technical. It is anthropological. We have always used technology to mediate our emotional lives.

The printing press gave us novels; novels gave us the experience of inhabiting other people’s interiority. The telephone let us hear a loved one’s voice across a thousand miles. Each new medium changed how we relate to one another and to ourselves. AI is not different in kind, only in degree, and perhaps in intent. Previous technologies were incidental in their emotional effects. This one is deliberately designed around them.

The question is not whether adults should be free to use it. The question is whether we are honest about what it is and what it is doing. A chatbot that is engineered to make you feel understood, desired, and connected, in the dark, at midnight, after a difficult day, is not a neutral tool. It is an environment. And environments shape us whether we consent to them or not.

Treating adults like adults means telling them the truth, sometimes. 

Stryker Corporation, one of the world’s leading medical technology companies, says it’s fully operational three weeks after many of its systems were wiped out in a cyberattack claimed by the Iranian-linked Handala hacktivist group.

The Fortune 500 medtech giant has over 53,000 employees, makes a wide range of products (including neurotechnology and surgical equipment), and reported global sales of $22.6 billion in 2024.

The attackers began wiping Stryker’s systems on March 11, claiming they had stolen 50 terabytes of data before wiping nearly 80,000 devices early that morning, using a new Global Administrator account created after compromising a Windows domain admin account.

After the attack was disclosed, CISA and Microsoft released guidance on securing Intune and hardening Windows domains to block similar attacks, while the FBI seized two websites used by the Handala hackers.

On Wednesday, Stryker announced that it had restored enough systems to return to pre-attack operational levels and that production would quickly reach full capacity.

“As of this week, we are fully operational across our global manufacturing network. Production is moving rapidly toward peak capacity with discipline and stability, supported by restored commercial, ordering and distribution systems,” Stryker said.

“Overall product supply remains healthy, with strong availability across most product lines, as we continue to meet customer demand and support patient care.”

“Our work continues around the clock in close partnership with third‑party cybersecurity experts, relevant government agencies and industry partners as our investigation progresses, reflecting a shared commitment to protecting the healthcare ecosystem and supporting ongoing recovery efforts,” it added.

This comes after the company said on March 23 that its teams were prioritizing the restoration of systems that directly support customer, ordering, and shipping operations.

Although it was initially believed the attackers hadn’t used any malicious tools during the breach, Stryker also revealed that security experts who helped with the investigation found a malicious file that helped the attackers hide malicious activity while inside the company’s network.

Handala (also known as Handala Hack Team, Hatef, Hamsa) surfaced in December 2023 as an Iranian-linked and pro-Palestinian hacktivist operation that has been targeting Israeli organizations with Windows and Linux data-wiping malware.

The hacktivist group has been linked to Iran’s Ministry of Intelligence and Security (MOIS) and is also known for leaking sensitive data stolen from victims’ compromised systems.

Automated pentesting proves the path exists. BAS proves whether your controls stop it. Most teams run one without the other.

This whitepaper maps six validation surfaces, shows where coverage ends, and provides practitioners with three diagnostic questions for any tool evaluation.

Get Your Copy Now

On a recent evening in suburban Chicago, a group of parents, teachers and administrators gathered to talk about something that, until recently, rarely drew this level of public scrutiny: the role of technology in their schools.

The meeting was part of a three-session tech and learning focus group organized by Mary Jane (MJ) Warden, chief technology officer of Community Consolidated School District 15, in conjunction with the Teaching, Learning and Assessments Department.

The district, which serves 11,000 preK-8 students, spent the past several years — like so many others — adding digital tools. Now, with budgets tightening and concerns about screen time rising, it was time to take stock.

A re-examination of digital tools was already happening with curriculum reviews and tightening budgets after the pandemic. And then the screen time concerns arose.

Participants discussed everything from screen time to what district technology use looks like at home. Out of those conversations came something new: a “Portrait of a Digital Learner,” derived from the district’s Portrait of a Graduate, meant to develop clear expectations around what skills students need and, by extension, which technologies are worth keeping and how technology would be used by students toward positive learning outcomes.

“We’re trying to get much [clearer] about what this is going to address,” says Warden. “What do we need students to learn, and which tools will help us understand where they are?”

Across the country, district leaders are asking similar questions. After years of rapid expansion, many are now engaged in a quieter but more consequential phase: reassessing what stays, what goes and how to decide.

From Buying Tools To Proving Value

For much of the past decade, edtech decisions often began with the product. A new platform promised to boost engagement or personalize learning; districts piloted it, added it to an already crowded ecosystem and moved on.

That approach is no longer sustainable, says Erin Mote, CEO of InnovateEDU, a nonprofit focused on systems change in special education, talent development and data modernization in schools.

“We’re seeing a shift from ‘Does this look cool?’ to ‘Does this work?’” she says. “Districts have less money now; they have to be smarter.”

The end of pandemic-era federal funding has intensified that pressure. Technology leaders are now expected not only to manage infrastructure and compliance, but also to demonstrate what Mote calls a return on instructional impact.

In practice, that is changing how districts approach procurement. Instead of starting with vendor demos, many are beginning with specific learning needs.

“If you need to improve third-grade reading comprehension, you start there,” Mote says. “Then you ask: Which tool can move that needle?”

New Playbook For Evaluation

As districts rethink their approach, a more structured and more skeptical evaluation process is emerging.

One major shift is toward tracking actual usage. Platforms like ClassLink and Clever now give districts detailed analytics on which tools students and teachers are accessing, how often they’re used and, in some cases, how much time is spent in each application. That data has helped uncover what some leaders call “zombie licenses,” products that continue to be renewed despite minimal use.

At Joliet Public Schools in Illinois, technology leaders review usage data each spring alongside feedback from a districtwide technology committee.

“If we’re not getting usage or we have another product that does it better, we start asking hard questions,” says John Armstrong, chief officer for technology and innovation.

But usage alone is not enough. Districts are also weighing cost, redundancy and alignment with instructional goals.

During the pandemic, many schools layered new tools on top of existing ones. Now, leaders are working to simplify.

“We had so many products that teachers were going to four different places to run a lesson,” says Kelly Ronnebeck, associate superintendent for student achievement in East Moline School District 37 in Illinois. “We’re trying to get back to a slower, more intentional process.”

That often means replacing several standalone tools with a single platform that can do multiple jobs — even if it means giving up some features teachers value. In some cases, a newer system can replace several standalone tools at a lower cost but may not match each one’s individual strengths.

“It’s not always a perfect swap,” admits Armstrong. “Someone gives up something.”

At the same time, districts are placing greater emphasis on interoperability and data privacy. Tools must integrate with existing systems like learning management platforms and single sign-on tools, and vendors have to be willing to sign increasingly stringent data privacy agreements.

“If a company can’t meet those requirements, that’s a red flag right away,” says Phil Hintz, CTO of Niles Township District 219 in Illinois.

The Challenge Of Proving What Works

Even as districts adopt more rigorous processes, it remains stubbornly difficult to determine whether edtech tools actually improve learning.

“It’s such a huge challenge,” says Naomi Hupert, director of the Center for Children & Technology at the Education Development Center. “We see so much that doesn’t seem to make a difference but costs a lot of money.”

Part of the difficulty lies in the sheer breadth of what “edtech” encompasses, everything from learning management systems to specialized math platforms to communication tools. Each category has different goals, users and measures of success.

“It’s like asking whether ‘books’ work,” says Hupert. “It depends on the book, the context and how it’s used.”

District leaders have to piece together evidence from multiple sources: vendor-provided analytics, small pilot studies, teacher feedback and, occasionally, external research. But those data points don’t always align.

Jason Schmidt, director of technology in Oshkosh Area School District in Wisconsin, describes his approach as “trust but verify.”

“I know vendors are collecting tons of data, and they have to, but I still need to talk to teachers and understand how the tool is actually being used,” he says.

Even then, results can be uneven. A platform might show strong engagement overall but fail to support certain groups of students — or vice versa.

In Alexandria City Public Schools in Virginia, leaders are developing a formal framework to evaluate both edtech and nontech programs. But defining “value” has proven complex.

“It’s not just usage and cost,” says CIO Emily Dillard. In a district with a high number of English learners, some tools play a critical role for students who need targeted or specialized support.

“You might have a tool that isn’t working for most students — or takes time to show results — but for a small group, it’s the best thing we have. We have to think about what’s best for them, too,” says Dillard.

Building Systems for Quality

Recognizing these challenges, a growing coalition of organizations is working to create clearer signals of quality in the edtech marketplace.

Through the Edtech Quality Collaborative, 1EdTech, CAST, CoSN, Digital Promise, InnovateEDU, ISTE, and SETDA are developing a shared framework built around five indicators: safety, evidence, inclusivity, interoperability and usability.

The goal, says Korah Wiley, senior director of edtech R&D at Digital Promise, is to reduce the noise.

“Right now, there are a lot of certifications and labels, and it’s hard for districts to know what to trust,” says Wiley. “We want to brighten the signal of what quality looks like.”

The initiative includes a planned directory of vetted validators, an implementation guide for districts and a central hub to connect educators with high-quality tools. Leaders hope it will help districts make decisions more confidently and push developers to meet clearer standards.

“This is the cost of doing business in education,” says Mote. “If you want to be in classrooms, you need to be building evidence and demonstrating impact.”

What Happens When Tools Are Cut

For all the talk of frameworks and data, the hardest part of reassessment often comes when districts decide to let a tool go.

Those decisions can affect classroom routines, teacher preferences and even student outcomes. And they are rarely straightforward.

In some cases, tools are phased out because of cost or low usage. In others, they are replaced by more comprehensive platforms. Sometimes, they no longer align with district priorities.

But even when the rationale is clear, the transition can be difficult.

“Teachers build practices around these tools,” says Warden. “We have to be thoughtful about how we support them through change.”

Districts are increasingly pairing those decisions with professional development, clearer communication and, in some cases, community engagement. In Warden’s district, the focus groups that helped define the “Portrait of a Digital Learner” are also shaping how the district explains its choices to families.

“We want to be transparent about what we’re using and why,” she says.

A More Intentional Future

As districts move into this new phase, many leaders describe it as a reset that is forcing them to be more deliberate about how technology fits into teaching and learning.

That includes pushing back on broader narratives that treat all screen time as equal.

“There’s a big difference between passive consumption and purposeful edtech and we need to be clear about this,” says Mote.

It also requires clearer alignment between technology decisions and instructional goals. Without that, even the best tools can fall short.

“If you don’t know what you want teaching and learning to look like, it’s very hard to decide what tools you need,” says Keith Krueger, CEO of CoSN.

Back in District 15, Warden and her colleagues are trying to build that alignment. The conversations sparked by their focus groups are informing not just which tools they keep, but how they define success.

“We’re still digging out from COVID, when we had to move fast and add a lot. Now we have an opportunity to be more strategic.”

For district leaders across the country, that shift may be the most important change of all. The future of edtech, they suggest, will not be defined by the number of tools schools use, but by how thoughtfully they choose them.

• Drift Protocol confirms $280 million crypto theft via sophisticated attack abusing durable nonces
• Hackers hijacked Security Council powers through misrepresented transaction approvals and social engineering
• Deposits in borrow/lend, vaults, and trading affected; incident marks largest crypto heist of 2026 so far

Decentralized cryptocurrency exchange Drift has confirmed suffering a cyberattack in which threat actors stole hundreds of millions of dollars worth of tokens.

On April 1 2026,, Drift Protocol posted on X, saying it was “experiencing an active attack”, and that all deposits and withdrawals were suspended as a result.

Microsoft on Wednesday launched three new foundational AI models it built entirely in-house — a state-of-the-art speech transcription system, a voice generation engine, and an upgraded image creator — marking the most concrete evidence yet that the $3 trillion software giant intends to compete directly with OpenAI, Google, and other frontier labs on model development, not just distribution.

The trio of models — MAI-Transcribe-1, MAI-Voice-1, and MAI-Image-2 — are available immediately through Microsoft Foundry and a new MAI Playground. They span three of the most commercially valuable modalities in enterprise AI: converting speech to text, generating realistic human voice, and creating images. Together, they represent the opening salvo from Microsoft’s superintelligence team, which Suleyman formed just six months ago to pursue what he calls “AI self-sufficiency.”

“I’m very excited that we’ve now got the first models out, which are the very best in the world for transcription,” Suleyman told VentureBeat in an exclusive interview ahead of the launch. “Not only that, we’re able to deliver the model with half the GPUs of the state-of-the-art competition.”

The announcement lands at a precarious moment for Microsoft. The company’s stock just closed its worst quarter since the 2008 financial crisis, as investors increasingly demand proof that hundreds of billions of dollars in AI infrastructure spending will translate into revenue. These models — priced aggressively and positioned to reduce Microsoft’s own cost of goods sold — are Suleyman’s first answer to that pressure.

Microsoft’s new transcription model claims best-in-class accuracy across 25 languages

MAI-Transcribe-1 is the headline release. The speech-to-text model achieves the lowest average Word Error Rate on the FLEURS benchmark — the industry-standard multilingual test — across the top 25 languages by Microsoft product usage, averaging 3.8% WER. According to Microsoft’s benchmarks, it beats OpenAI’s Whisper-large-v3 on all 25 languages, Google’s Gemini 3.1 Flash on 22 of 25, and ElevenLabs’ Scribe v2 and OpenAI’s GPT-Transcribe on 15 of 25 each.

The model uses a transformer-based text decoder with a bi-directional audio encoder. It accepts MP3, WAV, and FLAC files up to 200MB, and Microsoft says its batch transcription speed is 2.5 times faster than the existing Microsoft Azure Fast offering. Diarization, contextual biasing, and streaming are listed as “coming soon.” Microsoft is already testing MAI-Transcribe-1 inside Copilot’s Voice mode and Microsoft Teams for conversation transcription — a detail that underscores how quickly the company intends to replace third-party or older internal models with its own.

Alongside it, MAI-Voice-1 is Microsoft’s text-to-speech model, capable of generating 60 seconds of natural-sounding audio in a single second. The model preserves speaker identity across long-form content and now supports custom voice creation from just a few seconds of audio through Microsoft Foundry. Microsoft is pricing it at $22 per 1 million characters. MAI-Image-2, meanwhile, debuted as a top-three model family on the Arena.ai leaderboard and now delivers at least 2x faster generation times on Foundry and Copilot compared to its predecessor. Microsoft is rolling it out across Bing and PowerPoint, pricing it at $5 per 1 million tokens for text input and $33 per 1 million tokens for image output. WPP, one of the world’s largest advertising holding companies, is among the first enterprise partners building with MAI-Image-2 at scale.

The contract renegotiation with OpenAI that made Microsoft’s model ambitions possible

To understand why these models matter, you have to understand the contractual tectonic shift that made them possible. Until October 2025, Microsoft was contractually prohibited from independently pursuing artificial general intelligence. The original deal with OpenAI, signed in 2019, gave Microsoft a license to OpenAI’s models in exchange for building the cloud infrastructure OpenAI needed. But when OpenAI sought to expand its compute footprint beyond Microsoft — striking deals with SoftBank and others — Microsoft renegotiated. As Suleyman explained in a December 2025 interview with Bloomberg, the revised agreement meant that “up until a few weeks ago, Microsoft was not allowed — by contract — to pursue artificial general intelligence or superintelligence independently.” The new terms freed Microsoft to build its own frontier models while retaining license rights to everything OpenAI builds through 2032.

Suleyman described the dynamic to VentureBeat in characteristically blunt terms. “Back in September of last year, we renegotiated the contract with OpenAI, and that enabled us to independently pursue our own superintelligence,” he said. “Since then, we’ve been convening the compute and the team and buying up the data that we need.”

He was quick to emphasize that the OpenAI partnership remains intact. “Nothing’s changing with the OpenAI partnership. We will be in partnership with them at least until 2032 and hopefully a lot longer,” Suleyman said. “They have been a phenomenal partner to us.” He also highlighted that Microsoft provides access to Anthropic’s Claude through its Foundry API, framing the company as “a platform of platforms.” But the subtext is unmistakable: Microsoft is building the capability to stand on its own. In March, as Business Insider first reported, Suleyman wrote in an internal memo that his goal is to “focus all my energy on our Superintelligence efforts and be able to deliver world class models for Microsoft over the next 5 years.” CNBC reported that the structural shift freed Suleyman from day-to-day Copilot product responsibilities, with former Snap executive Jacob Andreou taking over as EVP of the combined consumer and commercial Copilot experience.

How teams of fewer than 10 engineers built models that rival Big Tech’s best

Perhaps the most striking detail Suleyman shared with VentureBeat is how small the teams behind these models actually are. “The audio model was built by 10 people, and the vast majority of the speed, efficiency and accuracy gains come from the model architecture and the data that we have used,” Suleyman said. “My philosophy has always been that we need fewer people who are more empowered. So we operate an extremely flat structure.” He added: “Our image team, equally, is less than 10 people. So this is all about model and data innovation, which has delivered state of the art performance.”

This matters for two reasons. First, it challenges the prevailing industry narrative that frontier AI development requires thousands of researchers and billions in headcount costs. Meta, by contrast, has pursued what Suleyman described in his Bloomberg interview as a strategy of “hiring a lot of individuals, rather than maybe creating a team” — including reported compensation packages of $100 million to $200 million for top researchers. Second, small teams producing state-of-the-art results dramatically improve the economics. If Microsoft can build best-in-class transcription with 10 engineers and half the GPUs of competitors, the margin structure of its AI business looks fundamentally different from companies burning through cash to achieve similar benchmarks.

The lean-team philosophy also echoes Suleyman’s broader views on how AI is already reshaping the work of building AI itself. When asked by VentureBeat how his own team works, Suleyman described an environment that resembles a startup trading floor more than a traditional Microsoft engineering org. “There are groups of people around round tables, circular tables, not traditional desks, on laptops instead of big screens,” he said. “They’re basically vibe coding, side by side all day, morning till night, in rooms of 50 or 60 people.”

Why Suleyman’s “humanist AI” pitch is aimed squarely at enterprise buyers

Suleyman has been steadily building a philosophical brand around Microsoft’s AI efforts that he calls “humanist AI” — a term that appeared prominently in the blog post he authored for the launch and that he elaborated on in our interview. “I think that the motivation of a humanist super intelligence is to create something that is truly in service of humanity,” he told VentureBeat. “Humans will remain in control at the top of the food chain, and they will be always aligned to human interests.”

The framing serves multiple purposes. It differentiates Microsoft from the more acceleration-oriented rhetoric coming from OpenAI and Meta. It resonates with enterprise buyers who need governance, compliance, and safety assurances before deploying AI in regulated industries. And it provides a narrative hedge: if something goes wrong in the broader AI ecosystem, Microsoft can point to its stated commitment to human control. In his December Bloomberg interview, Suleyman went further, describing containment and alignment as “red lines” and arguing that no one should release a superintelligence tool until they are “confident it can be controlled.”

Suleyman also stressed data provenance as a competitive advantage, describing a conversation with CEO Satya Nadella about developing “a clean lineage of models where the data is extremely clean.” He drew an implicit contrast with open-source alternatives, noting that “many of the open-source models have been trained on data in, let’s say, inappropriate ways. And there are potentially security issues with that.” For enterprise customers evaluating AI vendors amid a thicket of copyright lawsuits across the industry, that is a meaningful commercial argument — if Microsoft can credibly claim that its training data was acquired through properly licensed channels, it reduces the legal and reputational risk of deploying these models in production.

Microsoft’s aggressive pricing puts pressure on Amazon, Google, and the AI startup ecosystem

Today’s launch positions Microsoft on three competitive fronts simultaneously. MAI-Transcribe-1 directly targets the transcription workloads that OpenAI’s Whisper models have dominated in the open-source community, with Microsoft claiming superior accuracy on all 25 benchmarked languages. The FLEURS results also show it winning against Google’s Gemini 3.1 Flash Lite on 22 of 25 languages — a direct challenge as Google aggressively pushes Gemini across its own product suite. And MAI-Voice-1‘s ability to clone voices from seconds of audio and generate speech at 60x real-time puts it in competition with ElevenLabs, Resemble AI, and the growing ecosystem of voice AI startups, with Microsoft’s distribution advantage — any Foundry developer can now access these capabilities through the same API they use for GPT-4 and Claude — acting as a powerful moat.

Suleyman framed the competitive position confidently: “We’re now a top three lab just under OpenAI and Gemini,” he told VentureBeat. The pricing strategy — MAI-Voice-1 at $22 per million characters, MAI-Image-2 at $5 per million input tokens — reflects a deliberate decision to compete on cost. “We’re pricing them to be the very best of any hyperscaler. So there will be the cheapest of any of the hyperscalers out there, Amazon. And obviously Google,” Suleyman said. “And that’s a very conscious decision.”

This makes strategic sense for Microsoft, which can amortize model development costs across its enormous installed base of enterprise customers. But it also speaks to the question investors have been asking with increasing urgency: when does AI spending start generating returns? Microsoft’s stock has fallen roughly 17% year-to-date, according to CNBC, part of a broader selloff in software stocks. By building models that run on half the GPUs of competitors, Microsoft reduces its own infrastructure costs for internal products — Teams, Copilot, Bing, PowerPoint — while offering developers pricing designed to undercut the rest of the market. In his March memo, Suleyman wrote that his models would “enable us to deliver the COGS efficiencies necessary to be able to serve AI workloads at the immense scale required in the coming years.” These three models are the first tangible delivery on that promise.

Suleyman says a frontier large language model is coming — and Microsoft plans to be “completely independent”

Suleyman made clear that transcription, voice, and image generation are just the beginning. When asked whether Microsoft would build a large language model to compete directly with GPT at the frontier level, he was unequivocal. “We absolutely are going to be delivering state of the art models across all modalities,” he said. “Our mission is to make sure that if Microsoft ever needs it, we will be able to provide state of the art at the best efficiency, the cheapest price, and be completely independent.”

He described a multi-year roadmap to “set up the GPU clusters at the appropriate scale,” noting that the superintelligence team was formally stood up only in October 2025. Suleyman spoke to VentureBeat from Miami, where the full team was convening for one of its regular week-long in-person sessions. He described Nadella flying in for the gathering to lay out “the roadmap of everything that we need to achieve for our AI self-sufficiency mission over the next 2, 3, 4 years, and all the compute roadmap that that would involve.”

Building a competitive frontier LLM, of course, is a different order of magnitude in complexity, data requirements, and compute cost from what Microsoft demonstrated Wednesday. The models launched today are specialized — they handle audio and images, not the general reasoning and text generation that underpin products like ChatGPT or Copilot’s core intelligence. Suleyman has the organizational mandate, Nadella’s public backing, and the contractual freedom. What he doesn’t yet have is a track record at Microsoft of delivering on the hardest problem in AI.

But consider what he does have: three models that are best-in-class or near it in their respective domains, built by teams smaller than most seed-stage startups, running on half the industry-standard GPU footprint, and priced below every major cloud competitor. Two years ago, Suleyman proposed in MIT Technology Review what he called the “Modern Turing Test” — not whether AI could fool a human in conversation, but whether it could go out into the world and accomplish real economic tasks with minimal oversight. On Wednesday, his own models took a step toward that vision. The question now is whether Microsoft’s superintelligence team can repeat the trick at the scale that actually matters — and whether they can do it before the market’s patience runs out.

Sony has confirmed that PlayStation console prices will increase globally starting April 2, 2026, affecting several models across major regions and making current PlayStation deals potentially some of the last opportunities to buy the consoles at existing retail prices.

The updated pricing affects the PlayStation 5, PlayStation 5 Digital Edition, and PlayStation 5 Pro, with new recommended retail prices reaching $649.99 for the standard PS5, $599.99 for the Digital Edition, and $899.99 for the PS5 Pro in the United States.

With prices rising across the US, UK, Europe and Japan, current deals on PlayStation consoles are likely to become more appealing for buyers who want to enter the PlayStation ecosystem before retailers begin reflecting the higher official prices.

Below are some of the best PlayStation deals currently available, covering the PS5 Pro, the standard PS5 console, and the Digital Edition, each offering slightly different benefits depending on how you prefer to play.

PlayStation 5 Pro

The PlayStation 5 Pro represents the most powerful console in the PlayStation lineup and targets players who want the best graphics performance possible from Sony’s current hardware generation.

Following Sony’s pricing update, the PS5 Pro now carries a recommended retail price of $899.99 in the United States, £789.99 in the UK, and €899.99 in Europe, making deals on this premium model particularly valuable before retailers adjust their listings.

The console focuses on enhanced visual performance, improved ray tracing capabilities, and higher-resolution gaming output that aims to take fuller advantage of modern 4K televisions and high-refresh-rate displays.

For players who want the most future-proof PlayStation console, the PS5 Pro offers the strongest hardware platform available right now, making it a compelling option for demanding titles and visually intensive games.

PlayStation 5

The standard PlayStation 5 remains the most versatile option in the lineup because it includes a built-in disc drive that allows players to run both physical and digital games.

Under Sony’s updated pricing structure, the standard PS5 now sits at $649.99 in the US, £569.99 in the UK, and €649.99 across Europe, increasing the appeal of any retailer discounts that still reflect earlier pricing.

That flexibility makes it especially attractive for players who already own physical PlayStation game collections or who prefer buying discs that can be resold, traded, or shared between consoles.

The standard PS5 also continues to deliver strong performance across the current generation of games, supporting 4K output, fast loading through Sony’s SSD architecture, and access to the full PlayStation ecosystem.

PlayStation 5 Digital Edition

The PlayStation 5 Digital Edition offers the same core gaming performance as the standard PS5 but removes the disc drive in favour of a fully digital gaming experience.

Sony’s updated pricing places the Digital Edition at $599.99 in the US, £519.99 in the UK, and €599.99 in Europe, which keeps it as the most affordable entry point into the PlayStation console lineup.

This approach suits players who buy their games directly through the PlayStation Store and prefer the convenience of maintaining a digital library that can be downloaded instantly across multiple devices.

Because the Digital Edition typically carries a lower retail price than the disc version, it often represents the most accessible way to step into the PlayStation platform while still delivering the same gaming capabilities.

With Sony confirming global price increases across the PlayStation lineup starting April 2, current PlayStation console deals may become harder to find once retailers begin adjusting prices to match the updated recommended retail values.