Please support this podcast by checking out our sponsors:
- Invest Like the Pros with StockMVP - https://www.stock-mvp.com/?via=ron
- KrispCall: Agentic Cloud Telephony - https://try.krispcall.com/tad
- Lindy is your ultimate AI assistant that proactively manages your inbox - https://try.lindy.ai/tad

Support The Automated Daily directly:
Buy me a coffee: https://buymeacoffee.com/theautomateddaily

Today's topics:

Watermark removal versus provenance labels - An open-source tool claims it can remove visible and invisible AI image watermarks and strip provenance metadata, raising legal and trust concerns around media authenticity.

OpenAI adopts C2PA and SynthID - OpenAI is expanding image provenance with C2PA Content Credentials and adding SynthID watermarking, plus a verification tool—key moves for content authenticity and platform labeling.

How to evaluate AI agents - A new guide argues agent benchmarks must measure tool use, long-horizon reliability, and harness quality, using layered grading and outcome-based tasks like Terminal-Bench.

Open multimodal models get cheaper - Alibaba’s Qwen3 releases push multimodal and long-context capability into more efficient open models, lowering deployment costs for vision, video, and agent-style apps.

Pretraining surprises: mode-hopping - Researchers report “mode-hopping” during pretraining—models can abruptly switch between shallow heuristics and real reasoning—complicating how we pick checkpoints and data.

New efficient pretraining framework - Sapient’s HRM-Text open-sources a hierarchical recurrent approach and full training stack aimed at reducing compute barriers for from-scratch pretraining and reproducibility.

AI infrastructure, CPUs, and costs - NVIDIA’s Vera CPU begins shipping to major AI labs, while critics argue LLM economics remain shaky due to massive capex, power costs, and unclear AI revenue.

Censorship circuits inside model weights - A mechanistic interpretability study claims Qwen3.5’s PRC-related censorship is controlled by a small steerable circuit, making refusal behavior more legible—and manipulable.

AI backlash and courtroom drama - Graduates boo AI talk at commencements amid job anxiety, and a judge dismisses Musk’s lawsuit against Sam Altman on timing grounds, reshaping the OpenAI feud.

-Open-Source Tool Claims to Remove AI Watermarks and Provenance Metadata from Images
-Guide Explains How to Evaluate Long-Horizon AI Agents and Their Tool-Using Scaffolds
-Alibaba Qwen Releases Efficient Qwen3 Multimodal and Sparse MoE Models, Including FP8 Variants
-Study Finds Language Models ‘Mode-Hop’ Between Memorization and Generalization During Pre-Training
-Sapient Open-Sources HRM-Text, a Compute-Efficient 1B Language Model Pretraining Framework
-xAI Launches Grok Skills to Remember Workflows and Create Office Documents
-LLM Wiki v2 Proposes Lifecycle, Knowledge Graphs, and Automation for Durable LLM Memory
-Manus Scheduled Tasks 2.0 Brings Context-Aware Recurring Automation to Tasks, Projects, and Web Apps
-Anthropic Acquires SDK Tooling Company Stainless to Boost Claude Agent Connectivity
-Zenity Launches Three-Part Webinar Series on Enterprise AI Agent Security
-Cursor Releases Composer 2.5 With New RL Feedback and Larger-Scale Synthetic Training
-Commentary Warns AI Boom Economics Don’t Add Up for Clouds, Labs, or Customers
-NVIDIA shares LoRA/DoRA fine-tuning recipe for Cosmos Predict 2.5 to generate better robot manipulation videos
-Study Finds a Small, Steerable Censorship Circuit Inside Qwen3.5-9B
-Graduates Boo AI Talk at Commencements as Job Fears Grow
-Lovable Launches “Skills” to Reuse Task-Specific AI Instructions
-NVIDIA Starts Delivering Vera CPUs to Anthropic, OpenAI, xAI and Oracle Cloud
-OpenAI adopts C2PA and Google SynthID to strengthen AI content provenance and verification
-Zenity to Host AI Agent Security Summit 2026 in San Francisco
-Mistral AI Buys Emmi AI to Expand Physics-Based Industrial Engineering AI
-Algolia white paper outlines production blueprint for AI-powered search
-Judge Throws Out Musk’s Lawsuit Against OpenAI’s Sam Altman After Jury Finds It Was Filed Too Late

Episode Transcript

Watermark removal versus provenance labels
First up, the provenance arms race just got very real. A new open-source GitHub project called “remove-ai-watermarks” says it can remove both visible watermarks and invisible ones from AI-generated images, and also wipe out related provenance metadata. The headline feature targets Google Gemini’s visible “sparkle” mark, but the bigger claim is that it can also disrupt invisible schemes and strip fields that trigger “Made with AI” labels on major platforms. The repository does flag potential legal risk: in some jurisdictions, removing provenance with intent to deceive can cross into criminal territory. Why this matters is simple—industry and regulators are betting on labeling and traceability, and tools like this directly stress-test how durable those safeguards really are, whether for legitimate privacy reasons or for deception.

OpenAI adopts C2PA and SynthID
And that’s the perfect setup for the other side of the story: OpenAI says it’s expanding how it labels and verifies AI-generated media. The company is leaning harder into C2PA Content Credentials so platforms can read standardized provenance data more easily, and it’s adding Google DeepMind’s SynthID invisible watermarking to images generated through ChatGPT, Codex, and the OpenAI API. OpenAI is also previewing a public verification tool, basically a way to upload an image and check whether OpenAI-origin signals are present. The key point here is that metadata often gets stripped in the real world, while invisible signals can survive more handling—so OpenAI is using both. Taken together with watermark-removal tools, it’s clear we’re moving into an adversarial era for “what’s real” online, where signaling and signal-stripping will co-evolve.

How to evaluate AI agents
Shifting from media to agents: Cameron R. Wolfe published a detailed guide arguing that classic LLM benchmarks are no longer enough, because modern systems don’t just answer questions—they act over time, call tools, recover from errors, and operate inside messy environments. Wolfe’s main message is that you’re evaluating the whole setup: the model plus the harness around it, including instructions, tool access, and the system’s ability to manage context without drifting. He also pushes for layered evaluation—combining human review, deterministic checks where possible, and judge models carefully calibrated to humans. This matters because “agent performance” is quickly becoming a product claim, and without serious eval design, teams can mistake brittle demos for real reliability.

Open multimodal models get cheaper
On the practical side of making agents useful over the long run, a write-up proposing “LLM Wiki v2” argues that memory systems need a lifecycle, not just a pile of notes. The proposal emphasizes tracking confidence, handling superseded facts, and controlled forgetting so old information doesn’t drown out what’s current. It also advocates moving from flat pages to a typed knowledge graph, plus hybrid retrieval that mixes keyword search, vector similarity, and graph lookups. The significance here is governance and trust: as personal and team assistants accumulate months of context, the difference between “helpful memory” and “confidently wrong archive” becomes a make-or-break design problem.
<...