Identity, M&A Platforms, AI CoE, WebMD, IEEE Senior Member & Standards, EPM Columbia Business School EPM'26, Invited speaker - DOE Office of Science, IAAM'26.
Tweet 7/7:
Full paper: https://t.co/qqtDqzQQz6
Built on NIST SP 800-207 Zero Trust Architecture - ZTA secures identity at perimeter.
EHV constrains action within the system.
Identity + Action = complete governance.
Feedback welcome.
#AIGovernance#AgenticAI#HealthcareAI
Tweet 6/7:
The counterintuitive result:
Traditional assumption: ∂V/∂I < 0
(more governance = less velocity)
EHV result: ∂V/∂I ≥ 0
(governance co-located with execution = no tradeoff)
This sign reversal is the fundamental theoretical contribution.
Tweet 5/7:
Formal proof:
TLA+ Safety Invariant verified by TLC Model Checker.
Safety_TEE: Agent NEVER executes if TEE is compromised.
Safety_Policy: Partition forces ESCALATE state.
Non-compliant actions computationally unreachable.
Not monitored. Not flagged. UNREACHABLE.
Tweet 4/7:
EHV closes this gap.
Policy Enforcement Point relocated INTO the inference pipeline via Governance-Aware JIT Compiler.
G(a,C) ∈ {PERMIT, DENY, ESCALATE}
Evaluated pre-state. Not post-deployment. Not at audit.
AT THE MOMENT OF EXECUTION.
Tweet 3/7:
Every existing framework assumes humans are in the loop at speeds comparable to AI execution.
They are not.
NIST RMF: retrospective audit
ISO 42001: PDCA cycles
ZTA 800-207: perimeter only
None enforce constraints INSIDE the agent decision loop.
Tweet 2/7:
The problem in one sentence:
NIST AI RMF and ISO 42001 produce Governance Latency of 14–30 DAYS.
Agentic AI executes clinical decisions in microseconds.
That gap - 2×10⁹ milliseconds - is where patient harm lives.
Tweet 1/7:
New preprint live on arXiv.
"Ethical Hyper-Velocity (EHV): A Provably Deterministic
Governance-Aware JIT Compiler Architecture for Agentic Systems"
arXiv:2605.17909
🔗 https://t.co/qqtDqzQQz6
Thread 🧵
🚨 BREAKING: Someone just open-sourced software that sees you through walls using only WIFI signals.
it’s called WiFi-DensePose. It maps your exact body pose in real-time. no cameras. no sensors. just your living room router.
100% Open Source.
📢 Open-sourcing the Sarvam 30B and 105B models! Trained from scratch with all data, model research and inference optimisation done in-house, these models punch above their weight in most global benchmarks plus excel in Indian languages.
Get the weights at Hugging Face and AIKosh. Thanks to the good folks at SGLang for day 0 support, vLLM support coming soon. Links, benchmark scores, examples, and more in our blog - https://t.co/DcCG3zlN8p
Introducing Perplexity Computer.
Computer unifies every current AI capability into one system.
It can research, design, code, deploy, and manage any project end-to-end.
📡LIVE Now📡
Prime Minister @narendramodi participates in groundbreaking ceremony of India Chip Limited via video conferencing
Watch on #PIB's 📺
▶️Facebook: https://t.co/2NOan7mtux
▶️YouTube:https://t.co/Lm3Ai9J1Pm
https://t.co/7YySe63JDB
Interested in dialogue on scaling Human-centric AI to empower Healthcare and secure connectivity. Keen to see how India’s Digital Public Infrastructure (DPI) and "people-first" principles
Who’s attending? Let's connect! 🚀 #IndiaAIImpactSummit2026#IndiaAI#SafeAI#DigitalIndia