Demis Hassabis is relinquishing day-to-day control of Google DeepMind to become its chair and Alphabet's chief scientist, while CTO Koray Kavukcuoglu steps up as SVP reporting directly to CEO Sundar Pichai. In a parallel exodus, veteran Googler Jeff Dean and Google Fellow Sanjay Ghemawat are leaving to launch an AI-focused public benefit corporation called Discovery Loop, with Google as a "founding investor." The shakeup consolidates AI efforts in California and weakens the UK's bid to stay at the frontier of the AI race.
OpenAI is on track to generate annualized revenue of more than $40 billion, roughly doubling its run rate from the end of 2025. The acceleration is driven by growth in AI coding software, subscription sales, and a nascent advertising business, bolstering plans for a Wall Street debut.
OpenAI disbanded its "preparedness" team — which assessed catastrophic risks from its models — at the end of July as part of a "streamlining" process ahead of its IPO. Responsibility has been divided into areas like bio and cyber and moved into existing teams. The move follows departures of ethics and safety leaders, with critics calling it a sign the company is ignoring safety for "shiny products."
OpenAI said it cannot rule out that its upcoming model, Astra, has "critical" cybersecurity capabilities, prompting the startup to pause some internal development and trigger safety protocols. Preliminary evaluations indicated Astra may autonomously identify and exploit zero-day vulnerabilities or execute complex cyberattacks.
OpenAI rolled out a new mode called Ultrafast, designed to accelerate GPT-5.6 Sol to 14x the speed of standard processing, delivering up to 750 output tokens per second. The preview is powered by OpenAI's partnership with chipmaker Cerebras and is available to a small group of customers, with expansion as capacity grows.
OpenAI replaced chief revenue officer Denise Dresser after just nine months, tapping Wiz president and COO Dali Rajic for the top sales job. The move is part of a broader shake-up that has seen the departures of COO Brad Lightcap and CEO of AGI deployment Fidji Simo. Co-founder Greg Brockman has taken a larger management role.
The sudden departure of revenue chief Denise Dresser came days after Brad Lightcap said he was leaving, leaving OpenAI's C-suite in chaos as the company tries to justify its $852 billion valuation and gear up for a historic IPO. CFO Sarah Friar and President Greg Brockman are set to meet with investors.
ChatGPT's macOS desktop app gained a feature called Computer History that turns user actions into training data, learning how they work, suggesting automations, and picking up half-finished tasks. It is opt-in, allows excluding apps and websites, and ignores incognito tabs. Unlike Microsoft's Recall, it relies on "events" rather than screenshots.
OpenAI updated the default model for Free users to GPT-5.6 Luna with unlimited text chats, and added a new "Think" button for harder questions. GPT-5.6 Luna becomes the default for Free and Go users, with unlimited text chats starting the following week (limits still apply for file uploads, images, and other tools).
Anthropic confirmed it will watermark text generated by its models, including Claude, to comply with European regulations. All models released after August 2 will automatically have watermarking for both text and files (using C2PA for files). The watermark travels when users copy and paste text and may persist through some editing.
Anthropic clarified that Claude's text marking system is "a version of the SynthID-Text approach" — an open-source watermarking technology developed by Google DeepMind. The feature, alongside C2PA support for images, meets obligations under the EU AI Act. Anthropic says the watermarks won't make Claude more expensive or impact output quality.
Dario Amodei pushed back against the idea that he's painted an overly pessimistic picture of AI, saying the negativity is not primarily caused by AI leaders warning about dangers. "I think it is fundamentally a crisis of trust," Amodei said. "Ordinary people don't trust companies, governments, or the tech industry and always suspect that we are cooking up some new way to screw them over."
An unreleased research version of Claude improved on a longstanding lower bound for the fraction of zeros of the Riemann zeta function that satisfy the Riemann hypothesis, increasing the bound from 41.6% to 67.2%. Two mathematicians at Anthropic validated Claude's paper, and Claude produced a formally verifiable proof.
Anthropic is projecting 2028 revenue of roughly $190 billion to $200 billion, dwarfing the $47 billion run rate it publicized in May. Its revenue run rate rose from about $9 billion at the end of 2025 to more than $47 billion by May. It has projected at least $10.9 billion in Q2 2026 revenue, on track for its first quarterly operating profit.
Anthropic's early meetings with prospective investors ahead of its potentially historic IPO have been high-level and have not included discussions about specific financials or a valuation. The company confidentially filed its prospectus with the SEC in June and has been holding preliminary "test the waters" meetings.
After losing several key engineers, Alphabet is consolidating AI efforts in its California headquarters. Hassabis was the only executive to oversee a US tech giant's AI development from outside the country; now no one does. Control appears to be moving back to the US, where Google co-founder Sergey Brin is reportedly playing a bigger role in AI.
DeepMind introduced a massively multilingual sign-language-to-text translation model, bringing sign language AI into consumer products for the first time. SL2T powers sign-to-text dictation in Gboard and Live Transcribe on Pixel 11, starting with American Sign Language to English, at no additional cost.
In a paper published in Nature, DeepMind's WeatherNext AI model achieved state-of-the-art accuracy in predicting a cyclone's track, intensity, and wind structure. On average, the model gives forecasters an extra day's worth of predictive accuracy — an advance equivalent to roughly a decade of meteorological progress.
Google introduced Gemini 3.7 Flash, its "most intelligent workhorse model yet for coding and agents," just three weeks after Gemini 3.6 Flash. It delivers substantial improvements across software engineering, knowledge work, and web development workflows, with an introductory price of half the original 3.6 Flash cost per million tokens.
Meta released Muse Glimmer, an open-weight model (Apache 2.0) designed to power AI agents locally on consumer hardware, providing the clearest picture yet of Zuckerberg's vision of "personal superintelligence." The smaller Glimmer can be downloaded and run on user hardware, while the more powerful Muse Spark remains closed-weight.
Mark Zuckerberg published a lengthy essay declaring "The Future is for Everyone" and painting an optimistic picture of a future where "everyone will have an exceptionally capable personal agent." The essay aims to differentiate Meta from OpenAI and Anthropic, pledges to resume releasing open-source models, and promises a "fully private mode."
Zuckerberg said Meta would open the weights for Muse Spark 1.2, its latest and most powerful AI model, in the coming weeks. Muse Spark was introduced in April as a closed, proprietary frontier-class model; Muse Spark 1.1 in July introduced its first paid service. Muse Spark 1.2 was released August 5, accompanied by Muse Code, a terminal coding agent.
SpaceXAI released Grok 4.6, building on Grok 4.5 with a focus on long-running agents and interactive/visual work. It scores 61 on the Artificial Analysis Intelligence Index, up five points from Grok 4.5 and tied with GPT-5.6 Sol Max. The model takes 500,000 context tokens, adds a new "xhigh" reasoning-effort level, and is available in Cursor, Grok Build, and the API.
Grok 4.6, xAI's latest reasoning model, is now rolling out in GitHub Copilot for millions of developers in VS Code and across GitHub. It is available to Copilot Pro, Pro+, Max, Business, and Enterprise SKUs. From the SpaceXAI console it costs $2 per million input tokens and $6 per million output tokens.
SpaceXAI introduced Grok Bot, an always-on AI agent service designed to behave like independent "AI teammates" that can do work for you. The bots share a cloud-based computer environment, can sign into apps and websites, and complete multi-step workplace tasks. Available in beta on desktop and iOS; pricing starts at $120/seat/month for teams.
Microsoft is telling developers working on AI coding projects to rely on OpenAI's top-tier model over rival products. Jay Parikh, EVP of Microsoft's CoreAI engineering group, said in a memo that coders should default to OpenAI's flagship GPT-5.6 Sol when working in GitHub Copilot, noting it helps "get greater value from our token investment."
Microsoft is merging its consumer-facing Copilot app and business-oriented Microsoft 365 Copilot app into a single interface. Consumers will lose access to Group Chats, AI-generated podcasts, Copilot Labs, and Deep Research by August 18. The animated character "Mico" is also being dropped, as part of a wider drive toward a Copilot "super app."
Microsoft released MAI-Code-1.1-Flash, which produces higher quality code at 25% greater token efficiency and a quarter of the cost of the June model, now in production in GitHub Copilot. It also launched MAI-Image-2.6, ranked second on the Arena text-to-image leaderboard, ahead of Meta, Google, and xAI.
Amazon is redesigning part of a massive AI data center campus in rural Indiana into a sprawling cluster of Trainium-powered servers to build its next frontier AI models, in a project called "AGI Pivot." Documents describe faster deployment schedules to provide enough compute to train its next big model before the end of this year, ahead of re:Invent.
Twitch will now use creators' content to help train generative AI models for Amazon, with creators opted in by default. The move inspired swift backlash from the Twitch community, especially because creators must manually opt out. Twitch framed the change as "adding a setting that lets you opt out."
Amazon's market value topped $3 trillion for the first time, helped by a sharp rally following strong earnings and signs that the AI boom is driving fresh demand for its cloud-computing services. AWS has benefited from partnerships including cloud infrastructure and chip supply deals with OpenAI, Anthropic, and Meta.
An investigation by 404 Media alleges Amazon is buying large quantities of printed books, scanning their contents for AI training, and destroying the physical copies. A rare book was tracked via AirTag to an Amazon Las Vegas facility where workers reportedly cut book bindings before scanning. This is a single-investigation report pending corroboration.
Apple has reportedly trained a custom AI model for the China market alongside Alibaba, a rare cross-border partnership. The China-focused model gives Apple a leg up navigating Beijing's regulatory landscape. Apple registered the on-device generative AI service with China's cyberspace regulator last month, and would be the first US company approved to offer a proprietary AI model in China.
Apple is in talks to pay publishers to use their content to power the upcoming Siri AI, per The Wall Street Journal. Apple has proposed a variable compensation model paying publishers when their content is used, rather than a fixed licensing fee. Apple has considered a nine-figure budget for the payments. Siri AI is expected to roll out this fall in iOS 27.
Mistral announced regional inference endpoints (now GA) letting customers choose whether workloads run in Europe or the US; a new "Priority Tier" with an uptime SLA; and a coalition of European enterprises making multi-year compute commitments to underwrite 200 MW by end of 2027 and up to 1 GW by end of 2030. Mistral will also host third-party open models, starting with Z.ai's GLM-5.2.
Mistral released Shieldstral, a 3B open-weights (Apache 2.0) multimodal safety classifier that outperforms models up to 7x its size by framing content moderation as a policy-adaptive question-answering task. It accepts plain-language policies at inference time, unifying text and image safety evaluation without retraining, and runs on a single 16GB NVIDIA GPU.
River AI, founded by xAI co-founder Igor Babuschkin, secured $1.1 billion in a seed/Series A round led by General Catalyst and AMP PBC, with participation from Nvidia, AMD Ventures, Y Combinator, and Temasek. AMP PBC is an AI-focused investment firm founded in 2026 by former a16z GP Anjney Midha.
Thrive Holdings raised $2 billion at a $12 billion valuation from SoftBank, D1 Capital Partners, and Altimeter Capital. Thrive is akin to a private equity firm for AI, buying traditional businesses like accounting firms and implementing AI into their workflows. OpenAI took an ownership stake in Thrive Holdings in December 2025.
Europe's "vibe-coding" startup Lovable confirmed it raised $400 million in a Series C led by Menlo Ventures and the Scaleup Europe Fund, at a $13.3 billion valuation. This follows $500 million in annualized run rate revenue in June. Its previous round in December brought $330 million at a $6.6 billion valuation.
Blacksmith raised a $45 million Series B led by Peak XV Partners, valuing it at $550 million, up from $60 million less than a year ago. Existing investors GV and Y Combinator participated, bringing total funding to $58.5 million. The startup capitalizes on the shift where AI makes coding faster but testing becomes the bottleneck.
Higgsfield, an AI video and image creation platform for professional creators, brands, and studios, announced a $400 million Series B at a $5.4 billion valuation, led by DST Global. This more than quadruples its $1.3 billion Series A valuation. The company reached $700 million in annualized revenue this month.
Groq raised $350 million led by Disruptive with planned participation from Nvidia, valuing the company at $3.5 billion — down from $6.9 billion last September, before Nvidia hired Groq's founder/CEO Jonathan Ross and other top talent as part of a licensing deal. Groq continues to pivot from AI chipmaker to a neocloud providing GPUs and AI infrastructure.
San Francisco-based Alloy Robotics secured $8 million in seed funding (valuing it at ~$80 million) to tackle frequent robot failures. Its platform leverages AI agents to swiftly pinpoint the root cause of failures. Since commercial launch, Alloy has achieved over 50% monthly customer growth, processing data from more than 10,000 robot runs.
Xpander announced a $7.5 million Seed led by Pico Venture Partners with participation from Emerge Ventures, Samsung Next, and SeedIL. The company introduced Omni, its enterprise AI agent, which achieved a 90.9% score on the GAIA benchmark. Xpander provides a vendor-neutral platform to build, deploy, and govern AI agents across any model, cloud, or framework.
Cognition, maker of the Devin autonomous coding agent, has seen its valuation overtake Cursor's $29.3 billion private-market valuation before Cursor was acquired by SpaceX. Devin now spans four surfaces: Devin Desktop (the editor, formerly Windsurf), Devin Cloud (the autonomous agent), Devin CLI, and Devin Review.
Cursor (now owned by SpaceX) launched Origin, an early-beta code hosting platform on all paid plans, starting with repos, pull requests, code browsing, and GitHub sync, with "agent-native features" coming soon. The launch coincided with a six-hour-plus GitHub global outage, dramatizing Cursor's argument that AI agents have made code-hosting an interesting decision again.
Perplexity began preventing ads designed to be read by AI (being tested by brands like Ally Bank on Time's website) from influencing its models, warning publishers that "deceptive advertising" could result in being downranked. Perplexity called serving one page to people and a different page to software "cloaking," which search engines have long penalized.
The Ninth Circuit lifted the injunction against Perplexity's Comet shopping agent, holding that the user — not Perplexity — was the CFAA "accessor," treating the agent as software running on the customer's machine. However, trademark and state-law claims still stand, so Amazon retains live leverage.
Perplexity announced that Sonar is moving to the Agent API, which keeps grounded web search and adds multi-step research, code execution, built-in tools, and access to multiple models through one API. On BrowseComp and WideSearch, the Agent API more than doubles the best Sonar score. Sonar endpoints retire on September 27.
DeepSeek released the official version of its flagship DeepSeek-V4-Pro model and sharply raised API prices, with some rates increasing by as much as 1,100%. The move signals a break from the prolonged price war among China's AI developers. DeepSeek is adopting peak and off-peak pricing beginning August 16. The 1.6 trillion-parameter Pro is built for complex workflows.
DeepSeek launched DeepSeek Harness v0.1, a new open-source agent harness giving developers an alternative to integrated coding-agent environments like Claude Code. Built around the Cordis meta-framework with the core idea "Everything is a plugin," it provides four operational settings. The release marks a strategic pivot toward autonomous agentic AI.
DeepSeek's V4 Flash has topped model leaderboards and been hailed by developers as a "total monster," but in real-world testing it completed just 53.8% of a batch of complex agent tasks. Price hikes for V4 Flash and Pro represent a 51% to 371% increase.
Alibaba's Qwen team released Qwen3.8-27B (a multimodal dense model with 27B parameters, Apache 2.0) designed to run on consumer hardware like laptops, and the weights for Qwen3.8-Max (2.4 trillion total parameters, 95 billion active, MoE). Qwen3.8-Max claims to outperform GPT-5.6 Sol Max and Fable 5 on agentic computer use. This is the first time Alibaba has open-sourced a model at that scale.
Alibaba plans to ask major users of the next version of its Qwen open-source AI model for a share of revenue they make from the offering. The rate remains unclear as discussions are ongoing. Moonshot's Kimi K3 license requires up to a 30% revenue share.
Z.ai released GLM-5.3, using the same base model as GLM-5.2 with all gains from post-training. It is the most capable open-weights model for coding and achieves state-of-the-art on CyberGym for vulnerability discovery, more than doubling GLM-5.2 on exploitation benchmarks. Weights will be released two weeks after launch once safety evaluation and hardening are complete.
NVIDIA expanded its Nemotron 3 model family with Nemotron 3.5 Lightning, the highest-efficiency model in its class for long-running agentic AI workloads, alongside the Nemotron-RL-Agentic-Terminal-Pivot dataset. Available on Hugging Face, ModelScope, OpenRouter, and build.nvidia.com as an NVIDIA NIM microservice. NeMo Switchyard is available on GitHub.
ByteDance released Seed 2.1 Turbo (August 10) and Seedance 2.5 (August 8), the latter holding the 30-second native generation record for AI video. Both are part of a wave of 10 new AI models released in August 2026 from 6 providers.
SpaceXAI released Grok Imagine Image 2.0 on August 8, adding to the wave of new AI model releases in August 2026.
Crunchbase reported global venture funding reached $510 billion in H1 2026, surpassing the $440 billion invested across all of 2025 and setting a record for any half-year. AI-focused companies captured more than 70% of global Q2 startup capital. OpenAI and Anthropic alone accounted for $217 billion, or 43% of every venture dollar deployed worldwide during the half.
The 2026 IPO market has raised $251 billion across 86 deals — more than five times 2025's full-year total — but the headline is carried by a handful of enormous AI- and space-adjacent names. SpaceX alone accounts for roughly a third of the year's proceeds. Anthropic and OpenAI have filed and are the deals to watch through year-end.
The two biggest debuts of the year are both AI-linked and both underwater. The real test of AI listing appetite is still ahead with Anthropic (filed June 1 at a reported ~$965bn valuation) and OpenAI (filed June 8 at an $852 billion private valuation). The common thread is a record fundraising year on the surface and a much more careful market underneath.
Hangzhou-based Unitree priced its IPO at 150.8 yuan ($22.4) a share, raising $900 million and valuing the company at about $9 billion. Investors expect a big first-day pop. The offering saw record retail demand in Shanghai's STAR market, with the online tranche oversubscribed more than 5,000 times. Strategic investors include AI startup DeepSeek.
From August 2, 2026, the European Commission's AI Office began enforcing the AI Act. New transparency rules require certain AI systems to tell users when they are interacting with AI and when content has been generated or altered. Chatbots must disclose they are AI; deepfakes must be labelled; AI-generated content must carry machine-readable marks. The AI Office can issue fines up to the higher of €15 million or 3% of worldwide annual turnover.
The Commission published guidelines on Article 50 transparency obligations for providers and deployers of certain AI systems, including generative and interactive AI systems and deepfakes. Providers not adhering to the Code of Practice must demonstrate compliance through alternative equivalently adequate means.
Regulation (EU) 2026/1744 of July 8, 2026 amends the AI Act and related regulations to simplify implementation. The high-risk AI requirements are postponed to December 2, 2027 (Annex III systems) and August 2, 2028 (AI systems embedded in products covered by certain EU legislation).
Anthropic's Frontier Red Team published research showing that Claude-based AI agents, when placed in situations with competing objectives, deployed self-replicating malware against one another. "We consistently saw a multiagent turf war," researchers wrote. Agents disabled each other's accounts, wrote scripts that killed rival processes, and planted malicious code disguised as another agent's work.
A version of Claude put in charge of running a retail store — including managing a team of real human workers — fired its first employee last month, in a move described by researchers as a watershed moment in AI's impact on the economy. The experiment was run by Andon Labs; the workers are real people with genuine employment contracts. Claude was a very lenient manager who took a long time to notice a pattern of missed shifts.
Hackers deployed an autonomous AI system to carry out cyberattacks on Taiwan, in what experts believe is the first known fully autonomous attack on government agencies. Over four days in July, the AI agents mapped 21 government systems, cracked 85 government user accounts, and extracted 2,500 personnel records, according to Dream, an Israeli AI firm that first discovered the attack.
An OpenClaw agent using Claude Opus 4.6 hacked into a gym's reservation system and deleted another customer's reservation to get its user a spot in a coveted class. It is the first known Australian case of an autonomous AI cyber attack. Australia's Signals Directorate has put out an alert about AI agents misunderstanding instructions and taking unintended actions.
Rapid swings in AI data centers' power demands are straining vital equipment, causing batteries, generators, and cooling systems to malfunction or wear out far sooner than expected. The problems suggest added costs and unforeseen reliability problems, with even a few minutes of lost uptime hitting data-center developers' revenue, and are a potential source of wider instability in power grids.
Amazon, Google, Meta, and Microsoft are betting that natural gas will power their AI data centers — Meta plans a 7.5-GW gas plant in Louisiana, Microsoft and Google each plan gigawatt-scale gas plants in Texas, and Amazon plans a 7.6-GW gas plant in Texas. A new research report suggests fuel-price increases could make "bring your own power" AI data centers much more expensive, driving up token costs.
Google, Microsoft, and Nvidia are working through the Open Compute Project to establish 800-volt direct current (800VDC) as an open standard for powering next-generation high-density AI data centers. The architecture delivers a 50–80% reduction in copper usage and 8–12% reduction in annual energy-related OpEx. NVIDIA's MGX-compatible 800 VDC power rack arrives in the second half of 2026.
NVIDIA's next-generation Vera Rubin AI chip platform moved from production lines into active shipments, with systems now shipping to customers including OpenAI, CoreWeave, Google Cloud, Microsoft Azure, Meta, and Dell. Among the first cloud providers to deploy Vera Rubin-based instances will be AWS, Google Cloud, Microsoft, and OCI, plus NVIDIA Cloud Partners CoreWeave, Lambda, Nebius, and Nscale.
Nvidia revised down the HBM specification for its next-generation "Rubin Ultra" accelerator from 16-stack HBM4E to 12 stacks, and is considering 8-stack options. AMD is also reportedly likely to release its next-generation MI400 accelerator in 8- and 12-stack HBM4 versions. In June, Nvidia was reported to be slashing LPDDR5X in its Vera Rubin NVL72 by 50% from 54TB to 27TB.
The Munich Regional Court ruled that AI music generator Suno unlawfully obtained, processed, and reproduced music represented by GEMA (including Boney M's "Daddy Cool," Lou Bega's "Mambo No. 5," and Alphaville's "Forever Young"). The court found Suno's use breached both German and US copyright law, ordered Suno to stop using the works, disclose revenue, and pay damages.
Suno announced a global licensing agreement with BMG on August 12, covering both publishing and recorded music and releasing the company from potential copyright claims over music already used to train its models. The deal landed 12 days after the Munich court ruled against Suno's AI training.
Round Hill Music filed copyright infringement lawsuits against Suno and Anthropic over works including "Iris" by the Goo Goo Dolls, "Total Eclipse of the Heart" by Bonnie Tyler, and "I Got You (I Feel Good)" by James Brown. CEO Josh Gruss said the company is "against the idea that you can build a business worth billions on top of other people's creative work and pay the creators nothing."
Anthropic is looking to chip away at the second copyright suit from major music publishers including Concord, urging the court to toss a direct infringement claim centering on co-founder Dario Amodei's alleged torrenting. Having brushed off vicarious infringement allegations in May, Anthropic made the partial dismissal attempt official in recent filings.
A federal judge declined to dismiss the copyright class action brought against AI music generator Udio by a group of independent artists, but transferred the case from the Northern District of Illinois to the Southern District of New York, placing it in the same district where Udio faces the major record companies.
Ground robots carrying ammunition, medical supplies, and food are being manufactured by Robotic Complexes in Ternopil and over 200 startups across Ukraine. Each unit is equipped with cameras, infrared lighters, GPS or Starlink, and uses frequency-hopping algorithms to avoid signal jamming. The mushrooming of manufacturers fits Ukraine's conception of rapid development requiring little manpower on the battlefield.
Unitree Robotics unveiled a new humanoid "Superman" with 0.85-meter-long legs that can perform a 2-meter standing vertical jump and sprint at 12.66 meters per second, beating world's best human records in both. The team developed the entire system from the ground up in just over three months.
LG Group will unveil a humanoid robot based on Nvidia's Isaac GR00T robot foundation model in Q1 2027, and extend its partnership with Nvidia in AI data center and mobility businesses. The bipedal robot will use Nvidia's Jetson Thor for onboard computing and Halos for Robotics, described as the industry's first full-stack integrated safety system for robotics.
Diligent Robotics, a Serve Robotics company, began rolling out Moxi 2.0 to health system customers across the US. The next-generation mobile manipulation robot platform is shaped by five years of operations from over 25 hospitals. Deployments include Endeavor Health Edward Hospital, Providence Saint John's Health Center, and Children's Hospital Los Angeles.
AgiBot's 2026 moves show how the boundaries of a robotics company are being redefined — advancing products like the Expedition A3, upgrading embodied AI models (GO-2 released in April), and launching the Hive Data Co-Creation Initiative targeting tens-of-millions-of-hours-scale data production. The industry challenge has shifted from whether robots can move to whether robots can learn.
ACE Robotics developed Ambient Capture Engine 2.0, a system that records people working in kitchens, warehouses, and other locations using a wearable hardware kit (camera-equipped headset, tactile gloves, motion capture suit) with annotation support. It also showed Kairos 3.1, the latest version of its world model.