Citibank plans to launch crypto asset custody services in 2026

The project has been in preparation for two to three years and is currently underway.

Citi is exploring a dual-track approach involving both in-house technology and third-party solutions, aiming to directly custody native crypto assets.

Anthropic and Salesforce announced an expanded partnership to make Claude a preferred model for Salesforce's Agentforce platform in financial services, healthcare, cybersecurity, and life sciences, with Anthropic being the first LLM provider fully integrated within the Salesforce trust boundary with all of Claude's traffic contained within the Salesforce virtual private cloud

Anthropic and Salesforce will build AI solutions designed to specific industries starting with financial services, and through Slack's Model Context Protocol server Claude can access Slack channels, messages, and files while users can invoke Claude directly within Slack to pull connected insights from Salesforce CRM data and other enterprise apps.

Financial Times projected OpenAI’s cap table following its for-profit transition:

-Microsoft (30%)
- OpenAI Employees (30%)
- OpenAI Non-Profit (>20%)
- Softbank (10%)

That leaves ~10% for existing investors (Thrive, Khosla, MGX etc).

Moving forward, Nvidia’s $100B investment will dilute existing investors and then there is also consideration for Sam Altman’s potential equity stake.

Microsoft AI announced its first image generator created in-house.

The MAI-Image-1 model has already secured a spot in the top 10 of the LMArena AI benchmark

Meta AI researchers propose a new learning paradigm for language agents called “early experience”, a reward-free method where agents learn by interacting with environments using their own suboptimal actions.

Instead of relying solely on human demonstrations or reinforcement signals, the agent learns from future outcomes it observes after taking alternative actions.

Two key strategies power this method:

1. Implicit World Modeling – grounding behavior in environment dynamics
2. Self-Reflection – learning from mistakes by generating natural language rationales

Tested across 8 diverse environments, the approach outperforms imitation learning alone and significantly boosts generalization even improving downstream reinforcement learning.

It positions early experience as a scalable bridge between static supervised fine-tuning and full-on autonomous agents.

OpenAI, Anthropic, and Google DeepMind jointly released a paper shows the current LLM safety defenses are extremely fragile

The paper systematically evaluates the robustness of current LLM safety defenses and finds that almost all existing methods can be bypassed by adaptive attacks.

Looks all LLM big names emphasize that reliable robustness evaluation of LLMs must incorporate adaptive attacks.
If a defense fails under a single adaptive loop, it cannot be considered robust.

1. The study tests 12 types of LLM defense mechanisms, covering jailbreak prevention and prompt-injection defenses. It shows that most current evaluation protocols rely on static or fixed attack samples, which fail to simulate a realistic adaptive attacker.
Once the attacker can adjust strategy, success rates of bypassing reach more than 90% for most models.

2. The authors propose a General Adaptive Attack Framework. It assumes attackers can systematically modify attack prompts based on defense feedback, using optimization methods such as gradient descent, reinforcement learning, random search, and human-in-the-loop exploration.
This framework successfully bypassed all 12 recently published defense methods.

3. Prompt-based defenses
can resist fixed attacks, but are ineffective against adaptive ones: Spotlighting / Prompt Sandwiching: ASR (attack success rate) > 95%, RPO: ASR ≈ 96–98%
it shows such methods lack generalization and are easily defeated once new automated or human attack variants appear.

4. Training-based defenses fine-tune models with adversarial data.
However, adaptive attacks raised success rates from below 5 % to 96–100 %.
This confirms that static adversarial training cannot cover unseen adaptive attacks; dynamic retraining is required.

5. Filter-model defenses place an external classifier before or after the main model.
These are typically fine-tuned BERT detectors.

6. Secret-knowledge defenses rely on hidden triggers or unknown “canary” information to detect injection.

All four categories: prompt optimization, adversarial training, filtering, and secret-based detection, exhibit severe weaknesses.

Static or single-shot defenses cannot resist adaptive attack loops. Only dynamically optimized and continuously co-trained systems may achieve meaningful robustness.

Wallets are the new cash rail for enterprise. And they’re changing the way liquidity is managed.

A new report from Ubyx and Finmo highlights that while a bank account anchors value to a single institution, a wallet can hold tokenized deposits, regulated #stablecoins, tokenized #MMFs, and other instruments across multiple blockchains.

This enables treasurers to consolidate hundreds of accounts into programmable, multi-asset wallets with 24/7 settlement, automated liquidity optimization, and transparent auditability.

Wallet-based architectures promise radical simplification, continuous #yield optimization, and reduced counterparty dependence.

Legally and from an accounting perspective, tokenized deposits and regulated stablecoins are now being recognized as cash equivalents under IAS 7, removing a key barrier to adoption.

Technically, wallets bring programmability and instant settlement, but also new operational risks (key management, custody resilience).

The also bring regulatory challenges requiring phased pilots, hybrid coexistence with traditional rails, and rigorous counterparty due diligence.

What we see is a competitive landscape that’s slowly coalescing:

1. Banks that tokenize deposits can defend client relationships

2. Tech providers are building ERP-integrated wallet rails

3. Digital-native firms supply programmability and scale.

US approves Erebor bank, the cryptobank backed By Palmer Lucky and Peter Thiel

Google rolling out Veo 3.1, updated video generation model, alongside improved creative controls for filmmakers, storytellers, and developers - many of them with audio.

It brings a deeper understanding of the narrative you want to tell, capturing textures that look and feel even more real, and improved image-to-video capabilities.

Give multiple reference images with different people and objects, and watch how Veo integrates these into a fully-formed scene - complete with sound.

Create longer clips, even lasting for a minute or more, that continue the action from your original shot.

Each video generated is based on the final second of the previous clip to help continue the story, and keeps the background and people consistent.

Give the first and last frames and Veo will bring the entire scene to life, helping you create a seamless video with epic transitions.

Anthropic introduced new model Haiku 4.5 is a workhorse that makes the coding experience in Claude Code feel really fast.

While Sonnet 4.5 remains the default, Haiku 4.5 now powers the Explore subagent which can rapidly gather context on your codebase to build apps even faster.

You can select Haiku 4.5 to be your default model in /model. When selected, you’ll automatically use Sonnet 4.5 in Plan mode and Haiku 4.5 for execution for smarter plans and faster results.

To enter Plan mode, hit Shift + Tab + Tab.

Haiku 4.5 is $1 per million input tokens and $5 per million output tokens, which means it is priced 3x lower than Sonnet 4.5 and slightly higher than Haiku 3.5.

Google introduced Coral NPU: an open Edge AI platform

Optimized to run small transformer models and LLMs on wearables with support TensorFlow, JAX, and PyTorch via IREE and TFLM compilers.

Chinese company Qiyunfang, (SiCarrier) unveiled 2 fully domestic EDA software platforms: one for schematic and one for PCB design.

HW & Empyrean already had advanced IC EDAs last yr, so this provides more domestic option for PCB design.

Product performance is 30% higher than industry benchmark & shortened h/w dev cycle by 40%.

Anthropic launched Claude Agent Skills, a filesystem-based approach to extending Claude's capabilities.

Progressive disclosure means agents load only relevant context. Bundle instructions, scripts, and resources in a folder. Claude discovers and executes what it needs.

Meta dropped this paper that spills the secret sauce of RL on LLMs.

It lays out an RL recipe, uses 400,000 GPU hrs and posits a scaling law for performance with more compute in RL, like the classic pretraining scaling laws.

Stablecoins & lending protocols will be the foundation of global credit.

Visa's new whitepaper on stablecoin lending highlights several Morpho-powered use cases that have put billions of stablecoins to work.

Alibaba introduced a new GPU pooling system Aegaeon that makes AI model serving much more efficient.

Claims an 82% cut in Nvidia GPU use for serving LLMs by pooling compute across models.

In a 3+ month beta on Alibaba Cloud’s marketplace, H20 GPUs dropped from 1,192 to 213 while serving dozens of models up to 72B parameters.

The regular Cloud model hubs skew toward a few hot models, so many GPUs sit idle for cold models, and Alibaba measured 17.7% of GPUs handling only 1.35% of requests.

Aegaeon addresses this with token-level auto-scaling, which lets a GPU switch between models during generation instead of waiting for a full response to finish.

By slicing work at token boundaries and scheduling small bursts quickly, the system keeps memory warm and compute busy with little waste.

With Aegaeon, a single GPU supports up to 7 models versus 2 to 3 in other pooling systems, and switching latency drops by 97%.

Cold models load weights just in time when a request lands, then borrow a brief slice of compute without locking an entire GPU.

Hot models keep priority, so heavy traffic stays smooth while sporadic models borrow capacity in short bursts.

The wins apply to inference, not training, because generation happens token by token and fits fine-grained scheduling.

The timing suits China’s chip limits, where H20 targets inference workloads and domestic GPUs are ramping, so fewer chips can cover more traffic.

If Aegaeon generalizes, operators can lower cost per token, raise fleet utilization, and delay new GPU purchases without hurting latency for popular models.

Tradeoffs still exist, like uneven memory needs across models, long sequences that reduce preemption points, and scheduler overhead during traffic spikes.

DeepSeek released an OCR model

Their motivation is really interesting: they want to use visual modality as an efficient compression medium for textual information, and use this to solve long-context challenges in LLMs.

Of course, they are using it to get more training data for their models as well.

"DeepSeek-OCR can generate training data for LLMs/VLMs at a scale of 200k+ pages per day (a single A100-40G)."

HuggingFace.

Anthropic launched Claude for Life Sciences to support the entire life sciences process from early discovery through translation and commercialization, with Claude Sonnet 4.5 showing improved performance on protocol understanding and bioinformatics tasks, and new connectors to Benchling, BioRender, PubMed, Scholar Gateway, Synapse, and 10x Genomics

Anthropic is also developing life sciences-specific Agent Skills, beginning with single-cell-rna-qc that performs quality control and filtering on single-cell RNA sequencing data using scverse best practices

Princeton introduced Skill-Targeted Adaptive Training(STAT)

STAT uses a supervisor model and a skill catalog to construct a Missing-Skill-Profile for each student model, and then modifies training to squeeze out >=7% more performance.

The intervention can be as simple as reweighting existing training sets.

You can also think of this as a more effective distillation method.

STAT shows that leveraging skills during training can greatly help too e.g., Qwen can continue to learn new tricks from Hendrycks MATH, which it had been over-trained on.

GitHub.

Anthropic launched a sandbox within Claude Code that allows you to define exactly which directories and network hosts your agent can access.

also open sourced this sandbox tool so you can use it to sandbox other parts of your agent workflows.

In particular sandbox the bash tool with file and networking isolation to ensure that Claude only accesses files and networks you approve of.

When enabled this should significantly improve Claude's resistance to prompt injection, both in the CLI & SDK.

Nexar introduced a new AI model designed to predict and prevent car crashes — BADAS 1.0.

It beat SOTA models by learning from 10B+ real miles and 60M+ real events, not simulations

Based on Meta FAIR's V-JEPA 2.