Research
Multimodal Thinking with Renderable Programs
The authors introduce SVGLM, a framework that uses SVG primitives to connect text and images in reasoning, using SVG both as an image description and as text instructions.
Published
Author profile
Bryan Oubaita
Creator and owner of Baitaphish, publishing practical work on cybersecurity, cloud and AWS security, identity, phishing defense, AI security, and agent governance.
About
Security work grounded in inspectable evidence
Bryan Oubaita created Baitaphish and publishes practical cybersecurity education and source-linked research.
His work focuses on cloud and AWS security, identity, phishing defense, AI security, and agent governance. Baitaphish separates original analysis from source records and makes provenance, review dates, automation, and limitations visible.
Focus areas
Topics
- Cybersecurity
- Cloud and AWS security
- Infrastructure security
- Identity security
- Phishing defense
- AI security
- Agent governance
Verified public profile
Elsewhere
Published on Baitaphish
Research and articles
Research
ARGUS: Role-Aware Event Knowledge Graphs for U.S. Employment-Discrimination Complaints
The authors present ARGUS, a source-grounded, role-aware pipeline for constructing document-level Event Knowledge Graphs (EKGs) from U.S. employment-discrimination complaints, representing events with participants, temporal relations, and causal relations.
Published
Research
Never Too Late for Force: Accelerating VLA Post-Training with Reactive Force Injection
LIFT is a force-aware post-training framework for pretrained VLA policies that adds contact reactivity while aiming to preserve the pretrained manipulation prior.
Published
Research
Reachability-Based Formal Verification of Graph Neural Networks with Node and Edge Features
The authors introduce GraphStar sets, which extend Star sets to represent uncertainty in node and edge features, and implement them in GNNV to support reachability analysis for GCN and GINE architectures.
Published
Research
Who Finishes the Job? A Study of Follow-Up Fixes and Commit Authorship on AI Coding Agent Pull Requests
The study asks how often merged AI-agent PRs receive follow-up fixes, who authors those fixes, and how merge-time signals differ between agent PRs with and without a later fix.
Published
Research
From Agent Output to Authorized Transition
The central argument is that a present record, authentic execution trace, passing test, or successful tool result does not establish satisfaction of current requirements, legitimate caller authority, or correspondence between a later consequential action and the reviewed artifact; observation and evaluation outputs remain candidate evidence whose usable properties depend on trusted producer capabilities.
Published
Research
LabFactory: Building and Evaluating Executable AI Labs
The paper asks whether an agent can assemble the resources and procedures a scientific task requires into a system that remains invocable after construction ends, shifting attention from development actions or a single answer to capabilities retained in the delivered artifact.
Published
Research
Soundness Checking of Taint Flow Models
The authors contribute an approach that pairs LLM-generated taint flow models with a soundness checker, rather than constructing every model through inter-procedural taint analysis.
Published
Research
Inverse Problems Conditioned on Observation Ensembles: Applications and Methods
The paper formulates EIP-II as posterior inference for an observation y when both the prior and the forward model are unknown, using training data and a new observation set Y drawn under the same forward model.
Published
Research
When Labels Are Scarce: An Oscillatory State Space Model for Vibration Diagnosis
DualRes combines aligned spectral resolutions with normalized selective oscillatory updates; the paper’s evaluation contribution uses recording-level splits and explicit duration accounting to distinguish labelled exposure, recording coverage, and operating-condition shift.
Published
Research
Epistemic-Probabilistic Model for Guarded Multi-Agent LLM Coordination
The paper targets the lack of explicit social-knowledge representations and protocol-governed coordination in many LLM multi-agent systems. It proposes EPLA, a neuro-symbolic coordination architecture, and epistemic lottery gossip models that add agent-indexed lottery weights to call histories. Its announced Guard-decidability scope is an explicit modal-depth-one source fragment, not unrestricted epistemic reasoning.
Published
Research
Listen to the Latents: Self-Correcting Speech Recognition in Large Audio Language Models Through Hidden-State Interactions
The proposed decoding strategy reuses a preserved pre-adaptation language model to guide a speech-recognition model during inference.
Published
Research
LexIssue: Benchmarking Legal Issue Identification in Chinese Civil Litigation
The work introduces a legally grounded hierarchy that combines free-form descriptions of disputed matters with structured legal categories.
Published
Research
When Optimization Becomes Manipulation: Defending Generative Search against Malicious Generative Engine Optimization
The proposed defense combines protected evidence selection with regulated source use to address malicious optimization across the search-to-generation pipeline.
Published
Research
Automated Deployment of Real-Time Tasks for Phased Execution on Scratchpad-Based Multicore Platforms
The work proposes a model-based deployment methodology that automates application adaptation for phased execution.
Published
Research
Compute-Bounded Security Assurance - Coverage, Verification, and Response under Resource Constraints
The inquiry distinguishes repeated success, distinct coverage, accepted evidence, resource use, service capacity, and operational protection when evaluating added inference effort.
Published
Research
Subagents vs Agent Skills: Executing Reusable Knowledge for Long-Horizon Agentic Tasks
The proposed alternative invokes skill packages through delegated agents instead of placing their instructions in the primary agent context.
Published
Research
Halo: Improving forecast accuracy through heteroscedastic estimation
The study asks whether estimating both location and scale in time-series forecasting can improve point-estimate accuracy.
Published
Research
Token Merging for Multilingual Speech Recognition: A Systematic Study Across Model Scale and Fine-Tuning
The work systematically evaluates inference-time token merging across a multilingual speech-recognition model family and examines its interaction with parameter-efficient adaptation in lower-resource settings.
Published
Research
A Lifecycle Cost Analysis of Smart-Contract-Coordinated Federated Learning Marketplaces
The evaluated cost structure is amortizing because substantial deployment expenditure is paid per hired trainer only at setup, while recurring expenditure is markedly smaller and becomes diluted over the federation lifetime.
Published
Research
Using Codebooks to Detect Cybercrime Topics in Text Narratives
The proposed contribution is a generalist-oriented prompting method that uses qualitative codebooks to detect cybercrime incidents in consumer narratives with pretrained language models.
Published
Article
AI Agent Observability Is a Security Control
Security telemetry must connect agent identity, tool requests, authorization decisions, resource effects, and recovery—not just model latency.
Published
Article
The Real Permission Boundary of an AI Agent
An agent's effective authority spans identity, delegation, tools, credentials, policies, resources, and downstream effects—not its prompt alone.
Published
Article
Cryptographic Exposure Maps: Preparing a Repository for Algorithm Migration
A useful cryptographic inventory connects algorithms to libraries, protocols, keys, certificates, data, owners, and replacement constraints.
Published
Article
CVSS Is Not a Patch Queue: A Better Prioritization Model
Use CVSS as severity evidence—not a remediation order—by adding exploitation, exposure, business context, and control evidence.
Published
Article
GitHub Actions OIDC Removes a Secret, Not the AWS Trust Problem
GitHub OIDC eliminates stored AWS access keys, but workflow claims, role trust, session permissions, and resource reach still define risk.
Published
Article
IAM Wildcards Aren't the Whole Problem
Effective AWS authority depends on principal, action, resource, context, and every applicable policy—not the visible wildcard count alone.
Published
Article
Prompt Injection Is an Authorization Problem
Prompt injection manipulates model behavior; authorization, mediation, and resource reach determine whether manipulation becomes a harmful action.
Published
Article
Why Security Scanners Must Report Coverage Gaps
A zero-finding scan is uninterpretable unless the scanner records what it analyzed, skipped, could not build, and does not support.
Published
Article
What Source-Linked Security Intelligence Actually Requires
A source URL is not provenance. Useful intelligence preserves record identity, version, retrieval, claim scope, transformations, and uncertainty.
Published
Article
Static Security Exposure Graphs: Why Findings Aren't Enough
Repository findings become more actionable when connected to identities, credentials, policies, resources, data, and reachable effects.
Published
Research
Capability × Authority × Reach
A provider-neutral systems-security framework for constraining autonomous AI agents through capability assessment, effective authority, bounded reach, exposure-path analysis, and runtime trajectory assurance.
Published
Research
Secure AWS Architecture: Decisions, Controls, Evidence, and Recovery
A decision model for multi-account AWS security: trust boundaries, identity, preventive and detective controls, evidence, containment, and recovery.
Published
Research
Threat Modeling AI Agents: Trust Boundaries, Controls, and Evidence
A vendor-neutral threat model for AI agents that separates model behavior from authorization and maps threats to controls, evidence, tests, and residual risk.
Published
Research
Threat Modeling That Produces Testable Controls
A practical method for turning system boundaries and abuse paths into owned controls, observable evidence, and repeatable tests.
Published
Research
Why Provenance Matters in Security Intelligence
Provenance turns a security claim into a traceable record of source, transformation, review, and intended use—without pretending lineage alone guarantees truth.
Published
Research
NIST SP 800-61r3: Incident Response and Cybersecurity Risk Management
NIST's current incident-response guidance integrates preparation, response, recovery, and continuous improvement across the six CSF 2.0 Functions.
Published
Research
Repo Brief: Open Policy Agent (OPA) — Policy-as-code for authorization and guardrails
OPA lets you centralize policy decisions (Rego) for authz and compliance across microservices, Kubernetes, CI, and more.
Published
Research
Repo Brief: SigmaHQ/Sigma (Portable Detection Rules)
Sigma is a detection-rule format that lets you write once and translate to multiple SIEM backends (Splunk, Sentinel, etc.).
Published
Research
SLSA: Supply-chain Levels for Software Artifacts (What to actually implement)
SLSA is a maturity model for build provenance and tamper resistance. The real value is disciplined build pipelines and signed attestations.
Published
Editorial approach
Separate evidence, interpretation, and automation
Baitaphish writing names its sources, records review dates, labels AI assistance, and states what the available evidence cannot establish. Corrections should preserve the history of a claim rather than silently replacing it.