Technology News: Latest Updates from 2026-04-19

This report is a summary of technology news from 2026-04-19 and is based exclusively on the original texts of the sources.

Amazon ECR Pull Through Cache with Referrer Sync

According to aws.amazon.com, Amazon ECR Pull Through Cache now supports automatic discovery and synchronization of OCI referrers. This means that image signatures, SBOMs, and attestations from upstream registries are automatically pulled into private Amazon ECR repositories.

Previously, when retrieving referrers through a repository with a matching pull-through cache rule, the referrers from the upstream repository were neither returned nor synchronized. Users had to manually list and retrieve the upstream referrers. With the new update, Amazon ECR Pull Through Cache now directly accesses the upstream repository on Referrers API requests and automatically stores the associated referrer artifacts in the private repository.

As a result, end-to-end workflows for verifying image signatures, SBOM discovery, and attestation retrieval work seamlessly with pull-through cache repositories without requiring client-side workarounds. The feature is available in all AWS Regions where Amazon ECR Pull Through Cache is supported.

Cortex Code: Build dbt Projects Directly in the Terminal

According to snowflake.com, Cortex Code can scan source tables, generate models, add tests, run builds, validate results, and create a shareable report – all directly from the terminal. What previously required significant time and context-switching between various tools can now be accomplished faster.

Instead of starting with a blank SQL file, users can instruct Cortex Code to create a staging model for a table, convert timestamps, and rename columns to snake_case. Cortex Code understands dbt conventions and not only writes SQL but also automatically uses macros like {{ config(...) }} and {{ ref(...) }}.

This means the code is modular and “dbt native” from the start, making subsequent editing and optimization easier.

Fake Data Breach Alerts as a New Scam Tactic

According to eset.com, scammers are increasingly using fake data breach notifications to obtain personal information. In the US alone, 3,322 data breaches were reported last year, with nearly 280 million notifications sent to affected individuals. In Europe, daily incidents increased by 22% in 2025 to an average of 443 per day.

The scammers employ two tactics: They either exploit a real data breach and send fake messages in the wake of the news, or they fabricate a fictitious data breach of a well-known brand. Using AI tools, they create deceptively authentic messages in the respective local language, complete with logos and branding. The goal is to trick recipients into clicking malicious links, such as to install malware or steal personal data like Social Security Numbers.

For protection, the source recommends verifying suspicious notifications directly with the affected company – not through links in the message itself. Strong, unique passwords stored in a password manager along with multi-factor authentication (MFA) make it harder for attackers to gain access. A reliable antivirus and email security solution can help detect and block such phishing attempts early.

Samsung Receives Four Awards at Edison Awards 2026

According to news.samsung.com, Samsung Electronics was honored with two Gold and two Silver awards at the Edison Awards 2026, held on April 15 and 16 in Fort Myers, Florida. The awarded products – Smart Modular House, Vision AI Companion, Spatial Signage, and Bespoke AI Laundry Combo – were recognized for their innovative technological capabilities.

“Design stands at the intersection of business, technology, and humanity,” said Mauro Porcini, Chief Design Officer and President of the Device eXperience (DX) Division at Samsung Electronics. They will continue to drive innovation to enrich people’s lives in meaningful ways.

The Edison Awards were founded in 1987 by the American Marketing Association and named after inventor Thomas Edison. Winners are selected by an expert panel from industry, academia, and research based on rigorous innovation criteria. This year, awards were given in a total of 14 categories, divided into Gold, Silver, and Bronze tiers.

Nemotron OCR v2: Fast Multilingual OCR Model

According to huggingface.co, Nemotron OCR v2 is a multilingual OCR model designed for both accuracy and speed. It is based on 12 million synthetically generated training images in six languages. This reduced NED scores (Normalized Edit Distance) for non-English languages from 0.56–0.92 to 0.035–0.069. On a single A100 GPU, the model achieves 34.7 pages per second according to the source.

The predecessor Nemotron OCR v1 was a capable English OCR model but was not trained for multilingual purposes. It supported only 855 characters, which did not cover writing systems like CJK (Chinese, Japanese, Korean) or Cyrillic. An expansion to 14,244 characters brought only marginal improvements due to a lack of matching training data – the bottleneck was in the data, not the architecture.

The text source is mOSCAR, a multilingual web corpus with 163 language subsets covering scripts such as Latin, CJK, Cyrillic, Arabic, Devanagari, and Thai. The rendering pipeline is based on a heavily modified version of the Synthetic Document Generator (SynthDoG) from the Donut project. It produces pixel-precise annotations at three levels simultaneously – word-level, line-level, and paragraph-level – including reading order graphs. Various layout modes cover multi-column text, vertical text columns for Japanese and Chinese, tables, as well as PowerPoint-style and Word-document-style templates. The pipeline is language-agnostic and can be extended to any language for which fonts and source texts are available.

Conclusion

Today’s tech news highlights clear trends in AI-driven automation and security: From Snowflake’s Cortex Code, which builds dbt projects directly in the terminal, to Nvidia’s Nemotron OCR v2 achieving 34.7 pages per second on a single A100 GPU, to Samsung’s Vision AI Companion, AI is penetrating ever more product categories. At the same time, the threat from cybercrime is growing – scammers are using AI tools to create deceptively authentic fake data breach alerts, while AWS is improving security in container workflows with the new Referrer Sync for ECR Pull Through Cache.

Frequently Asked Questions

What improvement does the new Amazon ECR Pull Through Cache update bring?

Amazon ECR Pull Through Cache now supports automatic discovery and synchronization of OCI referrers. This means image signatures, SBOMs, and attestations from upstream registries are automatically pulled into private ECR repositories – without manual workarounds.

How do scammers exploit fake data breach alerts according to ESET?

Scammers either send fake messages in the wake of real data breaches or fabricate fictitious data breaches of well-known brands. Using AI tools, they create deceptively authentic messages complete with logos and branding. In the US alone, 3,322 data breaches were reported last year, while in Europe incidents increased by 22% in 2025 to 443 per day.

What performance improvements does Nemotron OCR v2 offer over its predecessor?

Nemotron OCR v2 was trained with 12 million synthetically generated training images in six languages. NED scores for non-English languages dropped from 0.56–0.92 to 0.035–0.069. Additionally, the character set was expanded from 855 to 14,244 characters to cover writing systems like CJK and Cyrillic.

📚 Sources

• 🔗 aws.amazon.com: Amazon ECR Pull Through Cache Now Supports Referrer Discovery and Sync…
• 🔗 snowflake.com: dbt Projects on Snowflake: Build & Deploy with Cortex Code…
• 🔗 eset.com: That data breach alert might be a trap…
• 🔗 news.samsung.com: Samsung Receives Four Accolades at Prestigious Edison Awards 2026…
• 🔗 huggingface.co: Building a Fast Multilingual OCR Model with Synthetic Data…