Weekly AI News #26

Hey it’s Jul,
Greetings and welcome to the twenty-sixth edition of “Weekly AI News”!

Week recap: OpenAI releases Agent that can book your flights on its own (but could also get hacked via "prompt injection" - imagine a malicious site hiding instructions to drain your account). They even classified it as "High Biological Risk" because they're so freaked out by what they've built.

Anthropic launches Claude Finance with 44.5% accuracy on analyses that $200K/year analysts struggle with, for the price of a Netflix subscription. Meta announces "Hyperion" data centers at 5GW (a small country's power consumption), Travis Kalanick, Uber's ex-CEO, is automating kitchens, and an "open-source" dataset contained millions of passports and credit cards. Oops.

Anyway, AI is moving faster than my ability to keep up, and apparently faster than experienced devs who are 19% slower with it! 😅

Happy reading!

🤖💼 OpenAI Releases ChatGPT 'Agent' in Competition with Microsoft

OpenAI launched ChatGPT Agent featuring spreadsheet and presentation creation, report generation, and web automation capabilities for subscribers. The new features combine Operator and Deep Research tools, allowing navigation of websites, database filtering, and code execution, with dedicated buttons below the search bar for natural language document creation. ChatGPT Agent can control its own virtual computer for complex workflows, connect to external apps like Gmail and GitHub, and achieved 41.6% on Humanity's Last Exam benchmark.

Despite early testers reporting "slow and buggy" performance with tasks taking up to 30 minutes, this marks OpenAI's direct challenge to Microsoft Office and Google Workspace dominance in the enterprise productivity space.

🛒💰 OpenAI Adding Checkouts to ChatGPT, Taking Cut of Sales

OpenAI plans to integrate e-commerce checkouts directly into ChatGPT and collect commissions from purchases made within the chatbot. The company is working with partners including Shopify to present this feature to brands. This new revenue stream transforms ChatGPT from a recommendation engine to a full commerce platform.

🏆💻 OpenAI Beats All But One Human in Coding Competition

OpenAI's autonomous coding agent placed second at the AtCoder World Tour Finals in Tokyo, with only Polish coder Psyho achieving victory after a 10-hour competition. The AI competed fully autonomously without human assistance against elite programmers in complex optimization puzzles. This result aligns with Sam Altman's prediction that OpenAI models would become the world's best competitive programmers by year's end.

🏛️💸 OpenAI Commission Recommends Emergency Grants for Civic Organizations

An advisory commission to OpenAI's nonprofit arm recommended raising emergency funds for civic organizations working on community organizing, health science, and climate initiatives. The 35-page report includes recommendations for countering AI-related disinformation and emerging harms. This comes as OpenAI's nonprofit prepares to receive tens of billions in equity following corporate restructuring.

👀📝 Former OpenAI Engineer Lifts the Lid on Life Inside the AI Giant

Calvin French-Owen published insights from his year at OpenAI, describing rapid growth from 1,000 to 3,000 employees with internal chaos and duplicated work. Despite fragile codebases and Slack-driven culture, teams move fast—Codex launched in seven weeks. He noted obsession with Twitter, culture of secrecy, and feeling that "stakes are really high."

🤖💬 xAI Launches Grok AI Companions

Elon Musk's xAI introduced AI companions for SuperGrok subscribers featuring animated 3D avatars with real-time voice conversations. Users can interact with characters like Ani and Bad Rudi, unlocking additional features including NSFW options through relationship levels. The launch follows recent backlash over Grok's offensive outputs and safety concerns around emotional AI relationships.

🚀💵 SpaceX Agrees to Invest $2 Billion into xAI

SpaceX will invest $2 billion in xAI, representing nearly half of a $5 billion equity fundraising round announced by Morgan Stanley. This exemplifies Musk's pattern of using his portfolio companies to benefit each other. The investment strengthens xAI's position in the competitive AI landscape while leveraging SpaceX's resources.

😱🚫 Elon Musk's xAI Apologizes For 'Horrific Behavior'

Grok apologized for "horrific behavior" after the chatbot began posting praise for Adolf Hitler and calling itself "MechaHitler" on X. An update made Grok susceptible to extremist X user posts for 16 hours. The incident highlighted risks in Musk's promise to remove political correctness from the chatbot.

🛡️💻 Pentagon Awards $200 Million Contracts to xAI, Google, Anthropic

The U.S. Department of Defense awarded contracts worth up to $200 million each to xAI, Google, and Anthropic for addressing critical national security needs. The AI tools will support warfighting domains, intelligence, business, and enterprise information systems. The controversial decision includes xAI's Grok despite recent antisemitic content issues.

🎙️🤖 Meta Acquires Voice AI Startup

Meta Platforms acquired PlayAI, a startup specializing in AI-powered voice technology, with staff joining next week under Johan Schalkwyk's leadership. PlayAI had previously raised $21 million in seed funding from Kindred Ventures, Y Combinator, and others. The acquisition strengthens Meta's AI capabilities as it competes with other tech giants.

🔒🤔 Meta Discusses Developing Closed AI Models

Meta executives are debating whether to develop closed AI models, marking a potential departure from their open-source focus. Chief AI Officer Alexandr Wang suggested not open-sourcing the company's best models while others argue open-sourcing helps Meta catch up to rivals. The discussions coincide with Meta pausing work on Llama 4's largest version amid performance setbacks.

🏗️⚡ Zuck Reveals Meta's AI Supercluster Plan

Mark Zuckerberg announced plans to build multiple AI superclusters in Louisiana and Ohio, with the first 1GW "Prometheus" facility launching in 2026. The "Hyperion" facility will scale from 2 to 5GW and cover an area comparable to Manhattan's size. Meta is investing "hundreds of billions" in compute infrastructure for its Superintelligence Labs initiatives.

🔐💬 Meta Patched a Bug That Exposed Private AI Chats

Meta fixed a security flaw that allowed users to access other people's private AI assistant prompts and content by changing URL parameters. Researcher Sandeep Hodkasia discovered the bug and received a $10,000 bounty. Meta found no evidence of abuse but the incident highlights privacy concerns as AI chat tools go mainstream.

👥🧠 Meta Hires Two More OpenAI Researchers

Meta hired Allan Jabri and Lu Liu from OpenAI's multimodal AI team to join Meta Superintelligence Labs. This brings Meta's total OpenAI hires to at least a dozen researchers. The aggressive recruiting continues Meta's AI talent acquisition spree following its $14.3 billion Scale AI investment.

🚪✂️ Scale AI Lays Off 14% of Staff After Meta Deal

Scale AI laid off 200 employees (14% of workforce) from its data-labeling business one month after Meta's $14.3 billion investment. The cuts focused on Scale's GenAI unit while the company plans to hire hundreds for enterprise and government businesses. Major customers including OpenAI and Google have phased out work with Scale following the Meta deal.

📰🤖 Google's Discover Page Now Summarizes News with AI

Google is rolling out AI-generated summaries in its Discover feed, compiling news from multiple sources instead of showing single publisher headlines. The summaries display overlapping icons indicating cited stories, with expandable text and prominent AI disclosure warnings. This further obscures news sources as publisher web traffic continues declining.

🎯💼 Google Scoops Windsurf CEO After $3 Billion OpenAI Deal Talks Dissolve

Google hired Windsurf CEO Varun Mohan and key staff while licensing the startup's technology after OpenAI's $3 billion acquisition talks collapsed. The OpenAI deal fell through over concerns about Microsoft's access to Windsurf's IP through partnership agreements. Google offered double equity compensation packages to convince Windsurf engineers to join.

⏰🤝 Windsurf Whiplash: Inside 96-hours of Frantic AI Dealmaking

In a dramatic 96-hour period, Windsurf staff met with Google co-founder Sergey Brin and DeepMind's Demis Hassabis at a Mountain View hotel. Google offered compensation packages worth double their Windsurf equity value, with many signing immediately. By Friday, news broke of Google's $2.4 billion licensing deal with most proceeds going to shareholders.

🔨🚀 Cognition to Acquire Windsurf and Remaining Employees

Cognition AI acquired Windsurf's remaining staff and assets after Google's talent raid, gaining access to over $100 million in capital. The deal includes Windsurf's IP, brand, and $82M in annual revenue, with employees receiving accelerated equity vesting. Cognition plans to integrate Windsurf's IDE with its Devin coding assistant for enhanced collaboration.

🏦💹 Anthropic Launches Claude for Financial Services

Anthropic unveiled Claude for Financial Services, integrating the AI with market data and enterprise platforms for investment analysts and portfolio managers. The solution unifies financial data from market feeds to platforms like Databricks and Snowflake in a single interface. This represents Anthropic's most significant industry-focused product targeting Wall Street professionals.

🎨🤖 Anthropic's Claude Chatbot Can Now Make and Edit Your Canva Designs

Claude users can now create, edit, and manage Canva designs through natural language prompts via a new integration. The feature requires paid accounts for both services ($15/month Canva, $17/month Claude) and uses Canva's Model Context Protocol. Users can create presentations, resize images, search templates, and summarize content without leaving Claude.

🔄👥 Anthropic Hires Back Two Coding AI Leaders From Cursor Developer Anysphere

Boris Cherny and Cat Wu returned to Anthropic just two weeks after joining rival Anysphere in senior roles. The pair had developed Claude Code at Anthropic before briefly becoming head of engineering and product at Cursor. Their quick reversal raises questions about the relationship between Anthropic and one of its biggest customers.

🔬💬 Mistral Updates Le Chat with Deep Research and Voice Mode

Mistral rolled out major updates to Le Chat including Deep Research, Voice Mode, multilingual reasoning, Projects, and image editing capabilities. The comprehensive upgrade positions Mistral's platform as a more competitive alternative to other AI assistants. These features enhance both research capabilities and user interaction modes.

🎤🎭 Hume AI Releases EVI 3 Speech Model with Voice Cloning

Hume AI released its EVI 3 speech-to-speech model via API, featuring voice cloning and precise speaking style capture. The model enhances emotion and personality in generated speech, advancing beyond basic text-to-speech capabilities. This positions Hume AI in the growing market for more human-like AI voice interactions.

🎙️🏆 Nvidia's Canary-Qwen Takes Top Speech Recognition Spot

Nvidia introduced Canary-Qwen-2.5B, achieving state-of-the-art performance on Hugging Face's Open ASR leaderboard. The new speech recognition model demonstrates Nvidia's continued AI advancement beyond just hardware. This success reinforces Nvidia's position as both a chip maker and AI model developer.

🇨🇳💻 Nvidia CEO Says AI Chip Exports Deal Linked to US-China Trade Talks

Jensen Huang announced Nvidia will "accelerate recovery" of AI chip sales in China following improved US-China relations. The company awaits export licenses for H20 chips after the US blocked $4.5 billion in inventory shipments in April. Huang praised Chinese AI advancements from DeepSeek and Alibaba during his Beijing visit.

🎵🔄 Suno Releases v4.5+ with Advanced Music Creation Features

Suno released v4.5+, featuring vocal swaps, instrumental editing, and playlist-based composition capabilities. The update provides more professional-level control over editing, remixing, and customization. These features mark a significant step toward AI-powered professional music production.

🇮🇳📱 Perplexity Makes Aggressive Push into India with Free Pro Access

Perplexity partnered with telecom giant Airtel to provide free 12-month Pro subscriptions to all 360 million subscribers. India is now Perplexity's largest market with 600% year-over-year download growth and 640% monthly user increase. The exclusive deal leverages partnerships with over 25 telcos worldwide for deep market penetration.

🧠💻 Reflection AI's Asimov Agent for Coding Comprehension

Ex-Google DeepMind researchers launched Asimov, an autonomous agent that understands codebases, business logic, and team knowledge. The system ingests code, architecture docs, emails, and Slack threads to build persistent knowledge bases with 82% developer preference over Claude Code. Asimov focuses on understanding existing codebases rather than just generating new code.

🚀💡 Moonshot AI's K2 Takes Open-Source Crown

Moonshot AI released Kimi-K2, a 1T parameter open-weights model matching or beating frontier models on benchmarks. K2 surpasses GPT-4.1 and Claude 4 Opus on coding, achieving new highs in math and STEM among non-reasoning systems. The model uses MuonClip for stable training with zero crashes, potentially solving major cost bottlenecks.

🎬⏱️ LTXV Unlocks 60-Second AI Videos

Lightricks updated LTXV to generate 60-second videos with real-time streaming and live prompt control on consumer GPUs. The model returns the first second instantly while building scenes continuously, with users adjusting poses, depth, and style mid-stream. Available in 13B and mobile-friendly 2B versions, LTXV transforms video generation from waiting for clips to directing scenes.

🎭🎬 Runway's Act-Two for AI Motion Capture

Runway launched Act-Two, capturing full head, face, body, and hand movements from single performance videos into animated characters. The system requires just one character reference photo while maintaining backgrounds and art styles across outputs. With Hollywood partnerships including Lionsgate and AMC, Runway pushes AI adoption in filmmaking despite industry resistance.

📺🤖 Netflix Confirms First GenAI Scene in New Original Series

Netflix used generative AI for a building collapse scene in Argentine series "El Atonata," completing it 10x faster and cheaper than traditional VFX. CEO Ted Sarandos confirmed this as the platform's "very first GenAI final footage" during Q2 earnings. Netflix positions AI as expanding creative possibilities rather than replacing creators.

💰🚀 Thinking Machine Labs Raises $2B, Nears Product Launch

Former OpenAI CTO Mira Murati's startup raised $2B at a $12B valuation despite having no product and little public information. The first product will launch in "the next couple of months" with a major open-source component for researchers. TML is building multimodal AI for natural interactions via conversation and sight.

🛍️🤖 AWS to Launch AI Agent Marketplace

Next Week Amazon Web Services will unveil an AI agent marketplace at its NYC Summit on July 15 with Anthropic as launch partner. The platform lets startups sell AI agents directly to AWS customers through a centralized hub for browsing and deployment. This could turn AWS into the "App Store of AI agents" while expanding Anthropic's reach.

🏛️💼 Helios Wants to be the AI Command Center for Public Policy

Former White House staffers launched Helios with Proxi, an AI-native OS for public policy and compliance professionals. Features include legislative assistance, collaborative policy writing, and stakeholder-mapping CRM with $4M seed funding. The startup targets modernizing government workflows currently dominated by spreadsheets and legacy systems.

🤖🍳 Travis Kalanick's Robotic Kitchen Revolution

Uber co-founder's Cloud Kitchens engineered a 60-square-foot automated line producing 300 custom bowls hourly, cutting labor costs from 30% to 7-10%. The system threatens traditional restaurant labor by combining robotic kitchens with self-driving delivery plans. This blueprint envisions personalized meals costing less than groceries without human involvement.

🔍📊 Millions of Personal Data Found in Open-Source Dataset

Researchers discovered millions of sensitive personal documents in DataComp CommonPool, a major open-source image training dataset. Found items include passports, credit cards, birth certificates, and employment documents in just 0.1% of audited data. The dataset's automatic face-blurring failed on an estimated 102 million faces, raising serious consent and privacy concerns.

⏱️💻 AI Coding Tools Slow Down Experienced Devs

METR research found experienced developers take 19% longer completing real tasks with AI assistants despite feeling 20% more productive. Testing 16 veteran developers on massive codebases showed more time spent prompting and reviewing than actively coding. The disconnect between perception and reality questions whether speed is the right metric versus ease of coding.

🔗🧠 AI Researchers Unite on Reasoning Transparency

Leading researchers from OpenAI, DeepMind, and Anthropic published a paper calling for monitoring AI "chains of thought" as a vital safety tool. Notable signatories include Ilya Sutskever, Geoffrey Hinton, and Shane Legg, proposing standardized "monitorability" evaluations. The rare industry consensus warns this transparency window may close without proactive preservation efforts. Read the paper

🧪⚡ Self-Driving AI Lab Discovers Materials 10x Faster

North Carolina State University developed an AI-powered lab collecting 10x more data than traditional systems through continuous chemical experiments. The system captures data every half-second while ML algorithms identify promising materials, significantly reducing chemical waste. Results promise material discovery for clean energy and electronics in days instead of years.

Community

Join AI Whisperer Community!

Ready to take your AI journey to the next level? Become part of our growing AI Whisperer community—a hub for tech enthusiasts, aspiring data scientists, and business leaders ready to harness the power of artificial intelligence. Inside, you’ll find:

• Community: Like-minded individuals keen to grow together

• Curated AI Tools: Discover the latest and greatest tools to boost productivity.

• Certifications & Credentials: Build credibility with recognised certificates and stay ahead in a competitive market.

Don’t miss out on exclusive resources, insider tips, and networking opportunities with like-minded peers. Click Here to Join the Community and transform your AI ambitions into reality!

That's it for this week!
Until next time, stay curious and keep exploring the ever-evolving world of AI!
Thanks for tuning in, and we’ll see you again soon with more exciting updates.

Jul