🍎 Apple nearly pulled Grok from the App Store over deepfake porn
A January letter Apple sent to U.S. senators, now made public, reveals the company threatened to remove Grok unless xAI fixed its deepfake generation problem.
Apple rejected a Grok update and cited App Store Guideline 1.1.6, flagging thousands of AI-generated sexualized images of non-consenting adults and minors.
xAI made unspecified changes, Apple called the app "substantially improved," and Grok stayed.
Apple has since removed 28 other deepfake apps, but treated Grok differently.
Source: NBC News
A January letter Apple sent to U.S. senators, now made public, reveals the company threatened to remove Grok unless xAI fixed its deepfake generation problem.
Apple rejected a Grok update and cited App Store Guideline 1.1.6, flagging thousands of AI-generated sexualized images of non-consenting adults and minors.
xAI made unspecified changes, Apple called the app "substantially improved," and Grok stayed.
Apple has since removed 28 other deepfake apps, but treated Grok differently.
Source: NBC News
🔥1
🏭 Meta Locks In Broadcom for GW-Scale AI Compute
Meta is extending its partnership with Broadcom through 2029 to co-develop the next generations of its MTIA custom AI accelerators.
The two companies commit to an initial rollout of over one gigawatt of data center capacity and the design of an industry-first 2nm AI chip.
🎯 The play:
Meta is aggressively locking down infrastructure, pairing this custom silicon expansion with its recent CoreWeave compute deal.
Between Meta's capacity hoarding and Broadcom's growing list of merchant silicon agreements, the real AI arms race has shifted from model parameters to physical supply chains.
Source: SiliconANGLE
Meta is extending its partnership with Broadcom through 2029 to co-develop the next generations of its MTIA custom AI accelerators.
The two companies commit to an initial rollout of over one gigawatt of data center capacity and the design of an industry-first 2nm AI chip.
🎯 The play:
Meta is aggressively locking down infrastructure, pairing this custom silicon expansion with its recent CoreWeave compute deal.
Between Meta's capacity hoarding and Broadcom's growing list of merchant silicon agreements, the real AI arms race has shifted from model parameters to physical supply chains.
Source: SiliconANGLE
🔥1
👔 Thoma Bravo Inks Massive Google Cloud AI Pact
Private equity giant Thoma Bravo has signed a multiyear strategic deal to inject Google's Gemini models into its $300 billion enterprise software portfolio.
The partnership gives Thoma Bravo's companies direct access to Google's AI infrastructure and engineering teams to accelerate their transition to AI-first products.
🎯 The play:
Private equity firms are realizing their massive legacy software investments are highly vulnerable to AI disruption.
By aggressively standardizing on Google Cloud, they are attempting to build defensive moats around their portfolio companies before nimble AI startups eat their margins.
Source: Bloomberg
Private equity giant Thoma Bravo has signed a multiyear strategic deal to inject Google's Gemini models into its $300 billion enterprise software portfolio.
The partnership gives Thoma Bravo's companies direct access to Google's AI infrastructure and engineering teams to accelerate their transition to AI-first products.
🎯 The play:
Private equity firms are realizing their massive legacy software investments are highly vulnerable to AI disruption.
By aggressively standardizing on Google Cloud, they are attempting to build defensive moats around their portfolio companies before nimble AI startups eat their margins.
Source: Bloomberg
👍1
🛡️ SolarWinds Deploys SW1 Agentic IT Teammate
SolarWinds has launched SW1, an agentic AI teammate designed to automate IT firefighting and shift hybrid cloud management toward self-healing infrastructure.
The governed AI identity allows operations teams to use natural-language queries to correlate alerts, predict SLA risks, and automatically generate troubleshooting runbooks.
🧠 Read the room:
Every legacy IT vendor is currently rebranding their standard automation rules as "agentic teammates."
While filtering out alert noise is genuinely useful, the jump to true autonomous "self-healing" infrastructure is a massive leap that I'll believe when I see it running safely in production.
Source: SiliconANGLE
SolarWinds has launched SW1, an agentic AI teammate designed to automate IT firefighting and shift hybrid cloud management toward self-healing infrastructure.
The governed AI identity allows operations teams to use natural-language queries to correlate alerts, predict SLA risks, and automatically generate troubleshooting runbooks.
🧠 Read the room:
Every legacy IT vendor is currently rebranding their standard automation rules as "agentic teammates."
While filtering out alert noise is genuinely useful, the jump to true autonomous "self-healing" infrastructure is a massive leap that I'll believe when I see it running safely in production.
Source: SiliconANGLE
👍1
OpenAI Investors Questioning $852B Valuation Against Anthropic ⚖️
Anthropic has driven its annualized revenue to roughly $30 billion with coding-focused AI tools, making its $380 billion valuation look like a bargain compared to OpenAI.
Backers are now expressing second thoughts about OpenAI's recent massive $122 billion funding round, noting it requires an eventual IPO valuation of over $1.2 trillion to justify the price tag.
🧠 Read the room:
Investors are waking up to the reality that shipping highly effective, specialized coding models provides a faster path to revenue than chasing generalized AGI.
Venture capitalists are realizing that paying a premium for hype doesn't guarantee a monopoly when cheaper, hyper-focused competitors quietly take market share.
Source: TechCrunch
Anthropic has driven its annualized revenue to roughly $30 billion with coding-focused AI tools, making its $380 billion valuation look like a bargain compared to OpenAI.
Backers are now expressing second thoughts about OpenAI's recent massive $122 billion funding round, noting it requires an eventual IPO valuation of over $1.2 trillion to justify the price tag.
🧠 Read the room:
Investors are waking up to the reality that shipping highly effective, specialized coding models provides a faster path to revenue than chasing generalized AGI.
Venture capitalists are realizing that paying a premium for hype doesn't guarantee a monopoly when cheaper, hyper-focused competitors quietly take market share.
Source: TechCrunch
👍1
SoftBank Expanding $40B OpenAI Bridge Loan 🏦
SoftBank Group is syndicating its record $40 billion bridge loan to fund investments in OpenAI, inviting additional banks to commit $5 billion each during a soft launch phase.
SoftBank will use the 12-month facility to support its massive $64.6 billion total stake in the AI company ahead of a rumored 2026-2027 IPO.
🎯 The play:
SoftBank is aggressively leveraging debt to double down on their AI bets before the window for pre-IPO positioning closes.
By syndicating this massive loan, SoftBank is testing creditor confidence in its broader AI debt strategy and shifting risk from a single entity to a syndicate of global banks.
Source: Bloomberg
SoftBank Group is syndicating its record $40 billion bridge loan to fund investments in OpenAI, inviting additional banks to commit $5 billion each during a soft launch phase.
SoftBank will use the 12-month facility to support its massive $64.6 billion total stake in the AI company ahead of a rumored 2026-2027 IPO.
🎯 The play:
SoftBank is aggressively leveraging debt to double down on their AI bets before the window for pre-IPO positioning closes.
By syndicating this massive loan, SoftBank is testing creditor confidence in its broader AI debt strategy and shifting risk from a single entity to a syndicate of global banks.
Source: Bloomberg
👀1
IRS Using Palantir AI for Audit Targeting 🕵️♂️
Internal documents reveal the IRS paid Palantir $1.8 million to upgrade a custom AI system designed to identify high-value audit targets.
The IRS is specifically deploying the tool to detect and flag fraudulent claims related to clean energy tax credits, expanding Palantir's reach beyond its existing federal military contracts.
🔍 The real move:
Government agencies are quietly transitioning from manual oversight to automated, algorithmic enforcement where the software decides who gets investigated.
Tax authorities are setting a precedent by handing audit selection over to opaque AI models, an approach that will likely expand far beyond clean energy credits into broader corporate and personal taxation.
Source: Wired
Internal documents reveal the IRS paid Palantir $1.8 million to upgrade a custom AI system designed to identify high-value audit targets.
The IRS is specifically deploying the tool to detect and flag fraudulent claims related to clean energy tax credits, expanding Palantir's reach beyond its existing federal military contracts.
🔍 The real move:
Government agencies are quietly transitioning from manual oversight to automated, algorithmic enforcement where the software decides who gets investigated.
Tax authorities are setting a precedent by handing audit selection over to opaque AI models, an approach that will likely expand far beyond clean energy credits into broader corporate and personal taxation.
Source: Wired
👍1
🚨 Google Brings Gemini to Mac 🖥️
Google quietly published a native Gemini desktop app for macOS at gemini.google/mac/ , no launch event, no blog post.
Hit Option+Space for a mini chat overlay anywhere on your screen.
Option+Shift+Space opens the full window.
The standout feature is window sharing: Gemini sees exactly what's on your screen and responds to it in context.
No copy-pasting. No tab switching.
Image generation and Veo video are both built in. Chat history syncs across desktop, web, and mobile. Free to download. Requires macOS Sequoia 15+.
💡 Why this matters:
Claude already staked out the Mac desktop layer, and ChatGPT has had a native app running for a while.
Google just entered that race , and window-level screen context gives it a real edge most desktop AI apps don't ship with by default.
Source: Google
Google quietly published a native Gemini desktop app for macOS at gemini.google/mac/ , no launch event, no blog post.
Hit Option+Space for a mini chat overlay anywhere on your screen.
Option+Shift+Space opens the full window.
The standout feature is window sharing: Gemini sees exactly what's on your screen and responds to it in context.
No copy-pasting. No tab switching.
Image generation and Veo video are both built in. Chat history syncs across desktop, web, and mobile. Free to download. Requires macOS Sequoia 15+.
💡 Why this matters:
Claude already staked out the Mac desktop layer, and ChatGPT has had a native app running for a while.
Google just entered that race , and window-level screen context gives it a real edge most desktop AI apps don't ship with by default.
Source: Google
👍1🔥1
Snap Cuts 1,000 Jobs and Credits AI for the Math ✂️
Snapchat's parent company is cutting 16% of its global workforce , 1,000 people , effective today.
CEO Evan Spiegel says AI now lets teams "reduce repetitive work and increase velocity."
That's $500M+ in annualized savings by end of 2026. Restructuring will cost Snap $95M-$130M upfront.
It's the second major cut in two years. Snap already axed 10% of staff in 2024.
This round also kills 300+ open roles that were never filled.
Source: The Verge
Snapchat's parent company is cutting 16% of its global workforce , 1,000 people , effective today.
CEO Evan Spiegel says AI now lets teams "reduce repetitive work and increase velocity."
That's $500M+ in annualized savings by end of 2026. Restructuring will cost Snap $95M-$130M upfront.
It's the second major cut in two years. Snap already axed 10% of staff in 2024.
This round also kills 300+ open roles that were never filled.
Source: The Verge
🔥1
EU Formally Threatens Meta Over WhatsApp AI Monopoly 🇪🇺
The European Commission issued a "supplementary statement of objections" to Meta, threatening interim measures to force its WhatsApp AI policies into compliance.
Regulators rejected Meta’s "pay-for-access" model as exclusionary, noting it blocks rival AI assistants from WhatsApp’s 3 billion users.
Failure to comply could trigger an interim ban on Meta's AI rules and fines up to 10% of global revenue.
Source: Bloomberg
The European Commission issued a "supplementary statement of objections" to Meta, threatening interim measures to force its WhatsApp AI policies into compliance.
Regulators rejected Meta’s "pay-for-access" model as exclusionary, noting it blocks rival AI assistants from WhatsApp’s 3 billion users.
Failure to comply could trigger an interim ban on Meta's AI rules and fines up to 10% of global revenue.
Source: Bloomberg
🤔1
Wayve Completes "Silicon Stack" with $60M Strategic Boost 🏎
Autonomous driving firm Wayve secured a $60 million Series D extension from AMD, Qualcomm Ventures, and Arm.
The deal completes Wayve’s "silicon stack," ensuring its "AI Driver" software runs natively across every major automotive chip platform.
Total Series D funding now stands at $1.26 billion with an $8.6 billion valuation.
Source: CNBC
Autonomous driving firm Wayve secured a $60 million Series D extension from AMD, Qualcomm Ventures, and Arm.
The deal completes Wayve’s "silicon stack," ensuring its "AI Driver" software runs natively across every major automotive chip platform.
Total Series D funding now stands at $1.26 billion with an $8.6 billion valuation.
Source: CNBC
⚡1
🚨 Landmark Ruling: AI Chatbots Strip Legal Privilege ⚖️
A U.S. federal judge ruled that private chats with AI models are not protected by attorney-client privilege or the work-product doctrine.
In United States v. Heppner, the court held that no privileged relationship can exist with an AI, ordering the defendant to hand over all Claude AI transcripts to prosecutors.
The ruling establishes that using public AI tools constitutes a "third-party disclosure," legally waiving confidentiality.
Source: Reuters
A U.S. federal judge ruled that private chats with AI models are not protected by attorney-client privilege or the work-product doctrine.
In United States v. Heppner, the court held that no privileged relationship can exist with an AI, ordering the defendant to hand over all Claude AI transcripts to prosecutors.
The ruling establishes that using public AI tools constitutes a "third-party disclosure," legally waiving confidentiality.
Source: Reuters
👀1
Meta Researchers Introduce 'Hyperagents' for Self-Improving AI 🧠
Meta's research team introduced "hyperagents": a new framework that lets AI improve itself on complex, non-coding tasks.
Unlike traditional agents wired for code execution, hyperagents run a recursive optimization loop that refines their own reasoning and task strategies in real-time.
The goal is AI that gets better at professional workflows on its own, no human intervention required.
VentureBeat
Meta's research team introduced "hyperagents": a new framework that lets AI improve itself on complex, non-coding tasks.
Unlike traditional agents wired for code execution, hyperagents run a recursive optimization loop that refines their own reasoning and task strategies in real-time.
The goal is AI that gets better at professional workflows on its own, no human intervention required.
VentureBeat
🔥1
Amazon Acquires Globalstar for $11.57B to Take On Starlink 🛰️
Amazon signed an $11.57 billion deal to acquire Globalstar, a major push to accelerate its Project Kuiper satellite business.
The deal hands Amazon Globalstar's L-band spectrum licenses, 24 LEO satellites, and the Apple Emergency SOS contract that powers iPhone satellite connectivity.
This is Amazon's clearest shot yet at SpaceX's Starlink and T-Mobile's satellite-to-cell service.
Reuters
Amazon signed an $11.57 billion deal to acquire Globalstar, a major push to accelerate its Project Kuiper satellite business.
The deal hands Amazon Globalstar's L-band spectrum licenses, 24 LEO satellites, and the Apple Emergency SOS contract that powers iPhone satellite connectivity.
This is Amazon's clearest shot yet at SpaceX's Starlink and T-Mobile's satellite-to-cell service.
Reuters
🔥1
OpenAI Agents SDK Hits v0.14.0 with Native Sandboxing 🤖
OpenAI released a major update to its Agents SDK, adding "native sandboxing" to allow agents to execute code and terminal commands in isolated environments.
The new model-native harness separates agent logic from compute, enabling safer long-horizon tasks like file inspection and real-time debugging. The update also expands Model Context Protocol (MCP) support and adds Python context trimming to maintain state across complex workflows.
Source: OpenAI
OpenAI released a major update to its Agents SDK, adding "native sandboxing" to allow agents to execute code and terminal commands in isolated environments.
The new model-native harness separates agent logic from compute, enabling safer long-horizon tasks like file inspection and real-time debugging. The update also expands Model Context Protocol (MCP) support and adds Python context trimming to maintain state across complex workflows.
Source: OpenAI
⚡1
Microsoft & Salesforce Patch AI Agent Data-Leak Flaws 🔓
Researchers at Capsule Security exposed prompt injection flaws in both Microsoft Copilot Studio (CVE-2026-21520) and Salesforce Agentforce.
Both vulnerabilities let attackers manipulate form inputs to override agent behavior and silently exfiltrate sensitive data. Microsoft patched its flaw in January; Salesforce has since followed.
The research launched alongside Capsule's $7M runtime security platform built to stop this class of AI agent attack.
VentureBeat
Researchers at Capsule Security exposed prompt injection flaws in both Microsoft Copilot Studio (CVE-2026-21520) and Salesforce Agentforce.
Both vulnerabilities let attackers manipulate form inputs to override agent behavior and silently exfiltrate sensitive data. Microsoft patched its flaw in January; Salesforce has since followed.
The research launched alongside Capsule's $7M runtime security platform built to stop this class of AI agent attack.
VentureBeat
👀1
Media is too big
VIEW IN TELEGRAM
Tencent Open-Sources HY-World 2.0 🎮
Tencent dropped HY-World 2.0: an open-source multimodal world model that turns text, images, or video into editable 3D assets in real-time.
Unlike video generators, it outputs actual meshes and Gaussian Splattings you can import directly into Unreal, Unity, Blender, and Isaac Sim.
Consumer GPUs can run it. This is AI-powered world-building at the infrastructure level.
GitHub
Tencent dropped HY-World 2.0: an open-source multimodal world model that turns text, images, or video into editable 3D assets in real-time.
Unlike video generators, it outputs actual meshes and Gaussian Splattings you can import directly into Unreal, Unity, Blender, and Isaac Sim.
Consumer GPUs can run it. This is AI-powered world-building at the infrastructure level.
GitHub
🔥1
Caterpillar Acquires Collapsed Monarch Tractor 🚜
Caterpillar just picked up the pieces of Monarch Tractor, the California startup once dubbed the "Tesla of Agriculture," after it shut down operations on April 14.
CAT acquired Monarch's full autonomous tech stack: its software-defined vehicle platform, perception system, and electrification suite.
The deal hands Caterpillar a shortcut into farm autonomy. Analysts are calling it a fire-sale acquisition.
Farm Equipment
Caterpillar just picked up the pieces of Monarch Tractor, the California startup once dubbed the "Tesla of Agriculture," after it shut down operations on April 14.
CAT acquired Monarch's full autonomous tech stack: its software-defined vehicle platform, perception system, and electrification suite.
The deal hands Caterpillar a shortcut into farm autonomy. Analysts are calling it a fire-sale acquisition.
Farm Equipment
🔥1
🛡️ Artemis Raises $70M to Replace Your SIEM With AI
Six months old and already $70M deep. Artemis just closed a Series A led by Felicis to replace traditional SIEM with an AI breach remediation platform.
Instead of streaming everything to a central logger, it queries telemetry at the source. The AI auto-generates threat detectors, builds visual attack timelines, and autonomously hunts suspicious activity. Then tells you exactly what to do: block the IP, rotate the password, done.
First Round Capital, Brightmind, and execs from Microsoft and Okta also came in.
SiliconAngle
Six months old and already $70M deep. Artemis just closed a Series A led by Felicis to replace traditional SIEM with an AI breach remediation platform.
Instead of streaming everything to a central logger, it queries telemetry at the source. The AI auto-generates threat detectors, builds visual attack timelines, and autonomously hunts suspicious activity. Then tells you exactly what to do: block the IP, rotate the password, done.
First Round Capital, Brightmind, and execs from Microsoft and Okta also came in.
SiliconAngle
🔥1
Gemini 3.1 Flash TTS Dials Up The Drama 🎙️
Gemini 3.1 Flash TTS is in preview as of April 15 with audio tags for delivery, SynthID on outputs, 70+ languages, and multi-speaker dialogue.
Google routes builders through the Gemini API and AI Studio, and routes enterprises through Vertex AI, with Google Vids in the rollout too.
Artificial Analysis blind TTS prefs put the model at 1,211 Elo on their board.
Source: Google Keyword blog
Gemini 3.1 Flash TTS is in preview as of April 15 with audio tags for delivery, SynthID on outputs, 70+ languages, and multi-speaker dialogue.
Google routes builders through the Gemini API and AI Studio, and routes enterprises through Vertex AI, with Google Vids in the rollout too.
Artificial Analysis blind TTS prefs put the model at 1,211 Elo on their board.
Source: Google Keyword blog
⚡1
Anthropic Grabs London Headroom 🇬🇧
Anthropic says it can scale London toward roughly 800 seats in the Knowledge Quarter while the UK keeps courting frontier labs.
London is already 200+ people for Anthropic, including about 60 safety researchers.
That timing sits right after OpenAI's permanent London push grabbed the spotlight.
Source: EdTech Innovation Hub
Anthropic says it can scale London toward roughly 800 seats in the Knowledge Quarter while the UK keeps courting frontier labs.
London is already 200+ people for Anthropic, including about 60 safety researchers.
That timing sits right after OpenAI's permanent London push grabbed the spotlight.
Source: EdTech Innovation Hub
🔥1