October 17, 2026

BerLabs — October 17, 2026

“thegentlemanpervert” producing AI CSAM is a NJ prosecutor, say cops

Brian Richard Fitzgerald, New Jersey's deputy attorney general, faces felony charges for producing and distributing AI-generated child sexual abuse material (CSAM). Authorities allege he transformed social media photos of real minors into sexualized images using AI and shared them online. This arrest follows intensified law enforcement efforts since 2024 to clamp down on AI-generated CSAM, reinforcing the legal stance that AI-created illicit content is subject to prosecution like traditional CSAM.

Why it matters
This case highlights escalating law enforcement actions against AI-generated CSAM, demonstrating that legal frameworks treat AI-produced abuse material as criminal, deterring misuse of emerging technologies.

Read the source →

Amazon wants you to Bee-lieve wearable AI can recall it all for you

Amazon is developing a wearable AI device designed to record and recall everything you experience throughout the day. This gadget aims to provide seamless memory assistance by capturing audio and possibly other inputs, letting users offload the mental burden of remembering details. While specifics on availability and privacy measures are still unclear, Amazon envisions this device as a new tool for personal productivity and memory augmentation.

Why it matters
If effective and privacy-conscious, this wearable AI could transform how people manage information and memory in daily life, shaping future productivity tools.

Read the source →

DeepSeek-v4.1-Flash-1017

DeepSeek-V4.1-Flash-1017 is a minor update enhancing multiple AI task performances including reasoning, coding, and visual benchmarks. The model reduces execution time and output length, improving speed and lowering operational costs. It supports various inference methods, programming languages, and deployment options, including Docker and APIs, aiming to advance the performance-budget balance for AI agents.

Why it matters
Faster, more efficient AI models like DeepSeek-V4.1-Flash-1017 enable broader, cost-effective deployment in real-world applications, benefiting developers and users by lowering barriers to AI-powered tasks.

Read the source →

pooling OpenAI subscriptions and bypassing regional inference capacity limits

Geoffrey Huntley released Underclass, a proxy pooling multiple ChatGPT subscriptions to efficiently manage inference requests and handle quota resets automatically. He also introduced Preflight, a proxy that scans and redacts sensitive data before sending requests to language models. Most notably, Huntley launched an experimental prototype called GeoClass that routes inference requests across globally distributed virtual machines, potentially overcoming regional capacity constraints seen in OpenAI's service, especially in Europe. These tools are composable and aim to improve performance amid the current compute shortage for large models.

Why it matters
As OpenAI users face throttling and capacity errors due to regional and global GPU shortages, these tools offer engineers practical ways to improve service reliability and performance by pooling resources and geo-routing requests.

Read the source →

All the news about the iPhone Duo

Apple announced the iPhone Duo, a foldable phone featuring a 5.4-inch outer and 7.6-inch inner display with a matte finish that reduces glare. It replaces Face ID with Touch ID on the power button, includes two cameras, and runs iOS 27 redesigned for foldables, including a side dock. Starting at $1,999 and available for preorder, it has already seen extraordinary demand, pushing delivery estimates to 2027, marking Apple's late but serious entry into the foldable market.

Why it matters
Apple's foldable debut signals a shift in premium smartphone designs and could influence software innovation across iOS devices.

Read the source →

Wall Street Week | America’s Relationship Woes, Battery Pivot, Farmers’ Fed Warning

At the Bloomberg New Economy Forum in New Delhi, global business and political leaders discussed realigning supply chains due to geopolitical risks. Former Canadian Deputy PM Chrystia Freeland and ex-US Treasury Secretary Janet Yellen argue in a new book that stronger US economic alliances are critical. Meanwhile, battery factories face challenges despite hopes for electric vehicle growth, with storage tech being explored as a potential solution. Rising US inflation is hitting the agricultural heartland particularly hard.

Why it matters
The US's economic future may depend on stronger global partnerships and adapting energy storage innovations amid inflation and supply chain disruptions.

Read the source →

Open-weight models win hearts, minds, and wallets

Open-weight AI models have sharply increased their market presence through Vercel's AI Gateway, processing over 70% of tokens by September and quadrupling spending share since June. This shift challenges traditional vendor lock-in, with nearly one-third of teams switching providers last month. While US firms like Anthropic and OpenAI still dominate spend, Chinese AI models' usage is rising, creating geopolitical uncertainty as China proposes new AI export controls. Experts warn that these developments complicate US businesses' reliance on Chinese models.

Why it matters
This trend reshapes AI consumption patterns, intensifies vendor competition, and raises strategic risks amid emerging US-China tech tensions, affecting where companies source AI services.

Read the source →

Continuous accountability for Verified bots

Cloudflare now continuously monitors Verified bots to ensure they respect website preferences and avoid abusive behavior. Beyond verifying bot identities, the system observes how bots follow site instructions, respond to feedback, and react to rate-limiting responses like HTTP 429. This creates a dynamic "trust band" reflecting a bot's consistent behavior over time, helping site owners manage bot traffic more confidently and allowing bot operators to prove their reliability.

Why it matters
This ongoing accountability system shifts bot trust from a one-time verification to a continuous evaluation, empowering site owners to better protect resources and enhancing responsible automation.

Read the source →