arXiv subjects

André Baptista

Publications and source records attributed to André Baptista.

3 recordsLinked to original sources

arXiv2026

From Controlled to the Wild: Evaluation of Pentesting Agents for the Real-World

AI pentesting agents are increasingly credible as offensive security systems, but current benchmarks still provide limited guidance on which will perform best in real-world targets. Existing evaluation protocols assess and optimize for predefined goals such as capture-the-flag, remote code execution, exploit reproduction, or trajectory similarity, in simplified or narrow settings. These tools are valuable for measuring bounded capabilities, yet they do not adequately capture the complexity, open-ended exploration, and strategic decision-making required in realistic pentesting. In this paper, we present a practical evaluation protocol that shifts assessment from task completion to validated vulnerability discovery, allowing evaluation in sufficiently complex targets spanning multiple attack surfaces and vulnerability classes. The protocol combines structured ground-truth with LLM-based semantic matching to identify vulnerabilities, bipartite resolution to score findings under realistic ambiguity, continuous ground-truth maintenance, repeated and cumulative evaluation of stochastic agents, efficiency metrics, and reduced-suite selection for sustainable experimentation. This protocol extends the state of the art by enabling a more realistic, operationally informative comparison of AI pentesting agents. To enable reproducibility, we also release expert-annotated ground truth and code for the proposed evaluation protocol: https://github.com/ethiack/ethibench.

cs.AI↗

arXiv2025

The substellar population in Corona Australis

The substellar initial mass function (IMF) and the formation mechanisms of brown dwarfs (BDs) remain key open questions in star formation theory. IMF characterization in a large number of star-forming regions (SFRs) is essential for constraining these processes. We aim to identify and spectroscopically confirm very low-mass members of the Corona Australis (CrA) SFR to refine its substellar census, determine its low-mass IMF, and compare it to other clusters. Using deep I-band photometry from SuprimeCam/Subaru and data from the VISTA Hemisphere Survey, we identified low-mass BD candidates in CrA. We subsequently obtained NIR spectra of 173 of these candidates with KMOS/VLT, as well as optical spectra for 8 kinematic candidate members using FLOYDS/LCO. The kinematic candidates are confirmed as low-mass stellar members with spectral types M1 to M5. In contrast, all 173 BD candidates observed with KMOS are identified as contaminants. Although the follow-up yielded no new substellar members, it places strong constraints on the number of undetected substellar objects in the region. Combined with literature data, this enables us to derive the substellar IMF, which is consistent with a single power-law slope of alpha = 0.95+-0.06 in the range 0.01-1 MSun or alpha = 0.33+-0.19 in the range 0.01-0.1 MSun. The star-to-BD ratio in CrA is about 2. We also provide updated IMFs and star-to-BD ratios for Lupus3 and ChaI from the SONYC survey, reflecting revised distances from Gaia. Finally, we estimate surface densities and median FUV fluxes for 6 SFRs and clusters and compare their substellar populations as a function of environmental properties. The IMF and star-to-BD ratio show no clear dependence on stellar density or ionizing flux from the massive stars. A combined effect - where one factor enhances and the other suppresses BD formation - also appears unlikely. (Abridged)

astro-ph.SR↗

arXiv2021

Pastprop-RNN: improved predictions of the future by correcting the past

Forecasting accuracy is reliant on the quality of available past data. Data disruptions can adversely affect the quality of the generated model (e.g. unexpected events such as out-of-stock products when forecasting demand). We address this problem by pastcasting: predicting how data should have been in the past to explain the future better. We propose Pastprop-LSTM, a data-centric backpropagation algorithm that assigns part of the responsibility for errors to the training data and changes it accordingly. We test three variants of Pastprop-LSTM on forecasting competition datasets, M4 and M5, plus the Numenta Anomaly Benchmark. Empirical evaluation indicates that the proposed method can improve forecasting accuracy, especially when the prediction errors of standard LSTM are high. It also demonstrates the potential of the algorithm on datasets containing anomalies.

cs.LG↗