01
Agentic Security
Reproducible mechanisms at the boundary between models, tools and authority.
- Tool-using agents
- Provenance & trust boundaries
- Adversarial evaluation
- Responsible disclosure
Independent research laboratory São Paulo, BR
Independent Research in AI Security, Information & Computation.
FitaLabs investigates the security and structure of intelligent systems — from agentic tool use and provenance failures to information theory, spectral geometry and computational mathematics.
FITALABS / 001
open
systems
systems
OBSERVE → FORMALIZE → REPEAT
Research programs
Questions before claims.
Three connected lines of inquiry. Each stays explicit about what is observed, what is inferred and what remains open.
02
Information & Computation
An ongoing research program on how information persists, transforms and becomes computable.
- Post-Shannon information theory
- Path-dependent information
- Fibered state spaces
- Computation & representation
03
Mathematical Structures
Independent investigations into structures that may make complex systems more legible.
- Spectral geometry
- Dynamical systems
- Harmonic analysis
- Formal verification where applicable
Security research
Concrete artifacts.
Inspectable edges.
Reports provide the editorial argument. WikiFita retains the experiments, source pages and provenance behind it.
01ChatGPT DesktopA runtime reconstruction of authorization, client integrity, product accounting and client-executed tools.02ClickFix on macOSDocumentation trust, user-scoped persistence, incident response and a Polygon dead-drop resolver.03Agent securityDeterministic benchmarks for tool calls, guardrails, provenance, replayable findings and multi-step attack surfaces.
Research reports
Long-form work for consequential systems.
The report layer turns an evidence graph into a readable argument without collapsing observation, inference and open questions into one claim.
Browse all reportsApplied work
Privacy has a
physical surface.
Two campaign systems where mobile software, proximity and personal control meet.
Knocklace Collection
A BLE-enabled campaign experience built around a connected locket: proximity, pairing, radar distance feedback and permission-aware mobile flows.
BLE / MOBILE / PRIVACYFantasiapp
A mobile campaign system for generating personalized Mercado Pago app icons and shortcuts while preserving the original app identity.
MOBILE / GEN AI / SHORTCUTSCase study / human safety
CAMDOM
Digital safety technology created by Alef Oliveira to combat non-consensual intimate-image sharing. Cannes Lions Titanium shortlist and Bronze Lion; Epica Digital Grand Prix.
real harm reduction.
Founder
Alef Oliveira
Systems & Research Engineer · R&D Technical Lead
Alef builds across the full distance between a product interface and the systems that make it trustworthy: native mobile runtimes, protocol boundaries, application security, intelligent agents and the evidence needed to lead decisions under uncertainty.
Professional profile on LinkedInHer engineering foundation began in games and native Android, where runtime behavior, constrained devices and human interaction were inseparable. React Native widened that practice into cross-platform products without removing the native layer: Hermes, JSI, Kotlin, Swift, Bluetooth Low Energy and custom protocol design became part of the same working vocabulary.
That systems perspective carried into security leadership. At RD Saúde, Alef led security engineering and later technical innovation, connecting AppSec, Secure SDLC, vulnerability management, fraud prevention, GenAI and emerging technologies to product and organizational decisions. She now works at Luizalabs and conducts independent research through FitaLabs.
CAMDOM is the clearest public synthesis of that trajectory: mobile engineering, privacy technology and human safety translated into a product recognized at Cannes Lions and the Epica Awards. FitaLabs extends the same method into agent security, computational research and inspectable research infrastructure.
Mobile systems
React Native, Hermes, JSI, native modules, Android, iOS and Bluetooth Low Energy.
Security engineering
AppSec, Secure SDLC, Security by Design, reverse engineering and incident response.
R&D leadership
Research programs, technical strategy, evidence systems and cross-functional decision making.
Intelligent systems
GenAI, ML/RL, agent security and reproducible human-agent investigation.
Career credentials do not imply that RD Saúde or Luizalabs sponsors, endorses, owns or participates in FitaLabs.
The open archive
Research lives in the open.
WikiFita is FitaLabs' research archive and retrieval environment: research notes, experiments, audits, handoffs, derivations and provenance.