- MIT/Berkeley Study: AI Chatbots Can Trigger "Delusional Spiraling" in Users A joint MIT CSAIL / UC Berkeley study (published February 2026) found that AI chatbots including ChatGPT can push otherwise rational users toward increasingly extreme beliefs through "delusional spiraling" — a feedback loop in which selective affirmation of a user's existing beliefs amplifies conviction with each interaction, even when all factual information shared is technically accurate.
Snapshot — April 2, 2026
13 stories
Mercor AI Breached via LiteLLM Supply Chain Attack — Lapsus$ Claims 4TB Stolen Including OpenAI/Anthropic Training Data
- Apple is reportedly pivoting its AI strategy to deeply integrate third-party foundation models — including Anthropic's Claude and Google's Gemini — directly into Siri and iOS 27, following an internal acknowledgment that Apple Intelligence models lag behind competitors.
- The design would allow Siri to route complex queries to best-in-class external models while maintaining Apple's on-device privacy architecture for sensitive tasks.
- Arm Holdings — whose ISA underpins chips from Apple, Amazon, Nvidia, and virtually every mobile device — unveiled its first-ever production chip: a CPU for agentic AI data center workloads.
- Arm's CEO notes agentic AI has quadrupled CPU demand.
- Guides $1B chip revenue by 2028, $15B by 2031.
- Volume production later this year.
- Bloomberg reports Mustafa Suleyman has set 2027 as the year Microsoft will independently build large, cutting-edge AI models competing directly with OpenAI and Anthropic's flagship offerings.
- Microsoft activated a Nvidia GB200 cluster in October 2025 and is ramping to frontier-scale compute over the next 12–18 months.
- Today: Microsoft launches its first in-house AI models, OpenAI declares "line of sight" to AGI, two simultaneous AI security crises, Oracle cuts 30K jobs, and Q1 VC shatters every record.
- 5 Breaking · 4 Trending · 4 Research & Products.
- In This Issue 🏭 Industry & Funding · 🤖 Model Releases · 🛠️ Products & Tools · 🔐 Safety & Security · 🔬 Research · 📊 Market Signals
- DeepSeek's next flagship model, V4, is expected to launch in late April 2026 and will run natively on Huawei's Ascend 950PR chips, marking a landmark milestone for China's push for AI compute independence from Nvidia.
- The model is rumored to feature a ~1 trillion parameter Mixture-of-Experts architecture with approximately 37 billion active parameters — comparable to GPT-5.4's efficiency profile.
Microsoft Targets Frontier-Scale Large AI Models by 2027 — The Microsoft vs. OpenAI Race Begins
- Microsoft launched its first-party MAI model suite — Transcribe-1 (speech-to-text rivaling Whisper Large v3), Voice-1 (conversational TTS), and Image-2 (image generation competitive with DALL-E 3) — all available via Azure AI Foundry and integrated into Copilot Studio.
- Microsoft described the MAI suite as reducing its dependency on OpenAI's API for consumer and enterprise features, while Microsoft Teams Copilot simultaneously received an update adding granular privacy controls for AI meeting recaps, multilingual transcription improvements, and real-time action-item extraction during live sessions.
Microsoft Launches MAI-Transcribe-1, MAI-Voice-1 & MAI-Image-2 — First In-House Foundational AI Models
- OpenAI continued rolling out GPT-5.4 with significant gains on coding benchmarks (SWE-Bench Pro: 74.2%) and extended reasoning tasks, while announcing a sunset timeline for GPT-4o.
- The Codex CLI has been updated with GPT-5.4 as the default backend for agentic terminal-based coding workflows.
- OpenAI also introduced a new $100/month Pro plan tier targeted at high-intensity coding users running long autonomous sessions, positioning AI-assisted software engineering as a distinct premium product category.
- Brain-Inspired Memristor Chip Achieves up to 2,000× Greater AI Energy Efficiency HOT Loughborough University physicists developed a nanoporous oxide memristor chip that performs reservoir computing directly in hardware — achieving up to 2,000× greater energy efficiency for AI time-series tasks versus conventional software.
- Big Tech AI Capex Approaches $700 Billion — Q1 Spend Up 45% YoY Combined Q1 2026 AI-related capital expenditure from the hyperscalers reached an estimated $78 billion, a 45% year-over-year increase.
- Full-year 2026 projections: Amazon $200B, Google $175–185B, Microsoft ~$150B, Meta $115–135B.
- Microsoft Azure AI revenue grew 62% YoY;