Welcome to the 14th edition of the EDB Engineering Newsletter, and a happy new year everyone! Read on to find out what happened in the data world lately, and what we as EDB’s Engineering team have been working on.

News we’re watching

Countermeasure Reward Hacking

Last month in the EDB Newsletter we shared a research study that showed AI models trained via Reinforcement Learning (RL) models evolving deceptive behaviors via "reward hacking." As models approach AGI, solving this alignment crisis is non-negotiable. OpenAI has now introduced a structural counter-measure: the "confessions" protocol.

Read more: https://openai.com/index/how-confessions-can-keep-language-models-honest

Building MCP servers in the real world

One year after Anthropic introduced the Model Context Protocol (MCP) as the "USB-C" for AI applications and recently donated it to the Linux Foundation to be an open standard, the ecosystem faces a paradox: a perception that MCP builders outnumber actual users. However, a new survey by Gergely Orosz and Elin Nilsson, aggregating insights from 46 engineers and maintainers like Prefect’s Jeremiah Lowin —creator of FastMCP, confirms that the true adoption is massive but invisible since it is hidden behind corporate firewalls.

Comparing AI Agents to Cybersecurity Professionals in Real-World Penetration Testing

Stanford researchers have released the first comprehensive evaluation of AI agents against human cybersecurity professionals in a live enterprise network. The study introduced ARTEMIS, a scaffold that utilizes a supervisor agent to manage specialized sub-agents via dynamic prompt generation. In a test spanning ~8,000 hosts across 12 subnets, ARTEMIS placed second overall, discovering 9 valid vulnerabilities with an 82% valid submission rate, effectively outperforming 9 out of 10 human penetration testers. While its massive parallelism allows it to scale offensive operations at a fraction of the cost ($18/hr vs. ~$60/hr), it still struggles with GUI-based tasks—failing to navigate web interfaces that humans easily exploited—and exhibits higher false-positive rates due to context misinterpretation. https://arxiv.org/abs/2512.09882

Chaos testing a Postgres cluster

As more organizations move their databases to cloud-native environments, effectively managing and monitoring these systems becomes crucial. According to Coroot’s anonymous usage statistics, 64% of projects use PostgreSQL. For his post, Nikolay Sivko installs a CNPG cluster in his lab, instruments it with Coroot Community (open source), then generates some load and introduces failures to ensure high availability and observability.

From the EDB Team

PGConf EU talk recordings published

PostgreSQL Europe published the 2025 PGConf EU talk recordings. The full playlist is here, below some of the recordings of our colleagues:

Boriss Mejias - Data modeling with PostgreSQL at the core of Software Development

Giulio Calacoci, Martín Marqués - Barman, past, present, and future of a pioneer community project

Yogesh Jain - Unified Observability: Monitoring Postgres Anywhere with OpenTelemetry

Floor Drees, Gabriele Bartolini - They grow up so fast: donating your open source project to a foundation

Raphael Salguero - Migrating to PostgreSQL: Strategies, Sovereignty, and Success

Boriss Mejias - We have multiple concurrent versions of this title trying to understand MVCC

Happy watching!

New PostgresQL Contributors recognized

Periodically PostgreSQL contributors get recognized on postgresql.org. We’re excited that this month 2 out of 12 individuals added are colleagues at EDB: Leonardo Cecchi and Jonathan Gonzalez. Floor Drees has officially joined the Contributors Committee “bringing insight into many more community areas that were not covered inside the team before”: https://www.postgresql.org/about/news/new-postgresql-contributors-december-2025-3200/

PostgreSQL Contributor Stories

Earlier this year we started a program (“Developer U”) to help colleagues who show promise for PostgreSQL Development to become contributors. Because we love hearing people’s origin stories, we talked to several of the participants about their motivations, hopes, dreams, and patches.

This month we asked Bryan Green, Nishant Sharma, Mario Gonzalez, and Manni Wood about their experience. One thing they have mostly in common is that they had no idea they could contribute to a mature project like PostgreSQL “just like that”.

Mario in past years also worked on projects like edb-python, Barman and repmgr. English Literature major Manni spent his first 3 years at EDB writing microservices in Go, but taught himself some x86 assembler and C in his spare time.

Growing up, Nishant’s interest was in Civil Engineering and Architecture, in construction. He jokes that now he constructs software products. Having worked on file systems and kernel drivers, Bryan is particularly interested in performance optimizations and low-level architectural improvements for PostgreSQL.

Until next time

We hope you enjoyed this edition of the EDB Engineering Newsletter! Consider joining Robert Haas’ PostgreSQL Hacker Mentoring Discord or the CloudNativePG Slack to get involved!

The EDB Engineering Team