Published track record

Numbers you can check.

Every result was published with Google or presented in public. Each names the DataPhant people involved and links to the source.

Google Research arXiv:2608.26088 · August 2026

Planetary Prediction Engine

An autonomous system that turns a plain-language question into a full geospatial model: it finds the data, screens it, fuses it with satellite and population embeddings, then trains and validates. Against expert pipelines it raised mean R² across 21 CDC health indicators from 60.0% to 76.8%, doubled accuracy on Nigerian food-security downscaling, and nowcast the 2026 DRC Ebola outbreak 10.3 points above the state of the art. Weeks of data engineering became minutes.

DataPhant authors · Jeremy Amez-Droz, Lambert Rosique Read the paper →
76.8% vs 60.0%
Black Hat USA Briefing · 2025

Autonomous timeline analysis and threat hunting for Timesketch

A digital-forensics agent built on Sec-Gemini for Timesketch, the open-source timeline tool. Instead of loading millions of log lines into a model, it decides what to check next, fetches only what it needs, and reconstructs the attack chain from initial access to lateral movement. On 100 real compromised systems it cut analyst triage time by more than 80%.

DataPhant authors · Lambert Rosique See the briefing →
>80% faster triage
Google Security Blog April 2024

Accelerating incident response using generative AI

Google’s incident responders spent up to an hour writing each incident summary. This work put a “Generate summary” button into their tooling, with a human editing every draft. In a blind test, security teams rated the LLM summaries 10% higher than human-written ones, and across 300 incidents writing time fell by 51%. Nothing was logged or retained.

DataPhant authors · Lambert Rosique, Jeremy Amez-Droz Read the post →
51% faster
Google Security Blog January 2024

Scaling security with AI: from detection to solution

LLMs wrote fuzz targets for more than 300 C/C++ projects in OSS-Fuzz, lifting coverage by up to 29% and exposing two new vulnerabilities in cJSON and libplist, libraries fuzzed for years. An automated pipeline then drafted and tested fixes for the bugs found, passing the best to a human reviewer. The framework was released as open source.

DataPhant authors · Jan Nowakowski Read the post →
+29% coverage
BuzzRobot Virtual talk · April 2024

AI to automatically fix security bugs

The vulnerability-fixing pipeline behind the Google paper. A Gemini-based model drafts patches for sanitizer errors in C/C++, Java and Go; patches are built and tested automatically, and only passing ones reach a reviewer. 15% of targeted bugs were fixed and merged this way, turning two hours of engineer time per fix into seconds.

DataPhant authors · Jan Nowakowski Watch the talk →
15% auto-fixed
Google DeepMind Gemini 1.5 technical report · 2024

Gemini 1.5: unlocking multimodal understanding across millions of tokens

The technical report for Gemini 1.5, built for long context. It retrieves a planted “needle” with over 99% accuracy up to 1M tokens of text, video and audio, and holds that recall out to 10M tokens of text, about 7M words. Jan and Lambert are listed contributors.

DataPhant authors · Jan Nowakowski, Lambert Rosique Read the report →
>99% recall

The same team, on your documents.

Request a demo