Omer Goldberg
@omeragoldberg
289 Following 11K Followers
Finance is the canary in the coal mine. Unfortunately, what kills the canary eventually kills everything else.
Finance is unforgiving, but it's not unique among enterprises adopting AI. Solving generally for finance opens the door to solving for many other industries.
So, more concretely, what is the problem at hand?
Agents and models don't know your business.
They don't know your
- enterprise context
- internal knowledge graph
- vertical regulations
- best practices, and more.
General agents haven't seen (or don't remember) your policies, your book, your incidents, or the hard-earned reasoning and insights your team has accumulated over the years.
Grabbing an API key and piping your data into an agent's context window is fine for a PoC. However, it's miles from a production-ready system. Finance is the canary because:
a) it's one of the largest enterprise markets in the world
b) the regulator is already in the room
c) the cost of being wrong can be quantified in real money, often on the same day.
What replaces demoware is the same wherever you build:
1) a structured reasoning layer beneath the model that actually encodes the entities, the mechanisms, and the institutional context that pretraining didn't see and that weights don't represent.
2) something real for the model to reason against, which can be your guardrails, deterministic risk harness, and internal benchmarks and eval sets.
Global markets move fast, so skipping this work shows its cost first.
Every other domain will follow the infrastructure, techniques, and harnesses built for finance
Great convo with @constkogan!
Show more
AI can sound smart.
Finance does not care.
@omeragoldberg breaks down why LLMs fail at the last mile, why static risk models cannot scale, and why the future of autonomous finance needs better risk infrastructure first.
Here’s my conversation with Omer Goldberg:
0:00 - Why on-chain risk is still misunderstood
1:19 - Omer Goldberg’s path from big tech to @chaoslabs
3:08 - The DeFi problem that made him start the company
6:19 - How Chaos Labs monetized risk intelligence early
8:54 - Why crypto keeps breaking under real stress
12:29 - What most people misunderstand about on-chain risk
16:10 - Why AI is powerful but still dangerous in finance
21:00 - The difference between crypto risk and traditional finance risk
24:46 - How nation state attackers think and operate
32:17 - Omer’s 3 biggest principles for risk management
37:05 - Why AI wrappers may not survive long term
42:14 - Where the biggest demand for financial AI is coming from
46:12 - Why risk infrastructure will matter more than hype
54:07 - The future of autonomous finance
59:38 - Omer Goldberg on the meaning of life
1:00:44 - The book that changed how he thinks
1:01:42 - Where to follow Omer Goldberg
Watch the full video on YouTube:
#OmerGoldberg# #ConstantinKogan# #ChaosLabs# #DeFi# #RiskManagement# #OnChainFinance# #AI# #CryptoSecurity# #Blockchain# #AutonomousFinance# #Fintech# #HolisticInvestments#
Show more
Great catching up with @constkogan
"Dump everything into a model and hope for the best" isn't a viable strategy.
From enterprises -> individual users: we're all internalizing this.
Ton of greenfield opportunity; enjoyed diving into what it means to build production-ready AI.
Show more
TradFi is adopting AI faster than crypto
🏛️ Most people assume on-chain finance will be the biggest buyer of “AI for finance.”
But the twist is TradFi may pull harder.
Not because they are more visionary, but because they are more constrained. Regulation forces higher standards, and in production, a small mistake is not just a bug. It can become a blow up.
⚖️ The part most people miss?
On-chain finance can tolerate “close enough” longer. Traditional markets can’t. That’s why purpose-built AI becomes a requirement, not a nice-to-have.
So the real question is not who talks about AI louder.
It’s who is forced to adopt it first.
📽️ Full Holistic Investment Podcast episode with @omeragoldberg , Founder and CEO of @chaoslabs, out tomorrow.
Watch on YouTube here!
#AI# #FinTech# #TradFi# #RiskManagement# #FinancialAI# #Regulation# #MarketStructure# #HolisticInvestmentPodcast# #ConstantinKogan# #OmerGoldberg#
Show more
Over the weekend, we identified an attack on Chaos Labs. The surface area was strictly contained to operational wallets we use for routine onchain operations. At no point was the Chaos Oracle Network breached or compromised. Chaos Oracles run in a fully isolated environment with nodes distributed globally, protected by layered security and cryptographic controls. The oracles continued to publish prices across every network throughout the incident.
We allocate a substantial share of our operating budget to cyber defense, alerting, and detection. These detection systems alerted us within seconds of suspicious activity, and we immediately moved to full lockdown.
Our incident response policy treats any threat as the worst-case scenario by default, given the value secured by the Chaos infrastructure. With the backdrop of recent exploits, we triggered our highest-severity incident response from the moment we were alerted and notified relevant partners. The authorities and cyber professionals working with us have characterized the activity as consistent with nation-state attacks.
As a precautionary measure, we rotated all keys, and have not observed any suspicious activity since.
The investigation continues, and we will share more as it allows.
Show more
Crypto is a swamp, with fewer than ~100 legit teams.
B/c of that, investing feels circular, with everyone in everything.
Marc was a savage; extremely fierce in pushing the DAO forward and always Aave first.
CT loves drama, but for anyone interested, it's all on the forums.
Show more
Since more than a decade, I've been supporting builders with small angel checks.
My thesis is simple: I'm good at generating money for both the projects I work on and myself, and I have a decent, low-profile lifestyle, with a reasonable house and a "boring" EV rather than mansions and supercars.
So part of my money is reinjected toward builders in verticals I want exposure to and want to support in their development.
- Most of these checks saw no return, money gone, it's part of the game.
- Some of them were quite profitable and bankrolled the next wave of investments
- And some yielded nothing, but I'm still proud of the product built.
is a good example. While as an investor I saw no return yet, I love their product suite and think @PhilippZentner is a chad.
Sometimes I bet on a team, sometimes I have a thesis and bet on a whole vertical hoping one of them ship a good product.
For example, a couple of cycles ago I was convinced MetaMask was a bad wallet that was hurting the global EVM user experience, so I invested in many alternatives.
When Rabby came out, I wasn't an investor. I still pushed them regardless because it was genuinely the best wallet. I think I contributed to its adoption in the ecosystem, and I'm proud of that, even if that meant no yield for me and less exposure to the ones I was invested in.
In the context of Aave, thinking a $20k angel check vested over years can influence an 8-figure position in a project I spent my sweat and blood on is a bit delusional.
My thesis on Aave has always been to carefully curate the best team in a vertical and contribute to kingmaking them. When you're managing onchain lending, you need to be opinionated, in terms of relationships and pure risk surface. One good king at $10B TVL is simply superior to 5 unequal guys at $2B each on average. This has yielded amazing results with Lido, Etherfi, Ethena (my pushback on Resolv saved Aave), Maple, and Pendle.
For Kelp, we onboarded them, after pushing back and voting against due to a weak oracle, locked up with only borrowing wstETH and with clear caps to generate an LRT <> LST <> WETH flywheel generating dozens of millions of dollars of revenue that is currently financing the paychecks of the people who seem less focused on "leaving no ghost behind", which was our mantra under my tenure, and more focused on trying to point fingers.
Expanding the scope of assets it was allowed to borrow is something I pushed back on for a while because it was conflicting with our existing relationships.
Liquid (re)staking in a protocol like Aave is a delicate balance with incredibly valuable users who hold their position literally for as long as the carry is profitable. Adding competitors to these guys can increase borrow costs, and throwing Lido users under the bus to benefit some LRT users might be a good short-term idea, but if not carefully tailored (as we did with Etherfi), it will eventually backfire.
Reality is more pragmatic. I do not pretend to own domain expertise on every topic, otherwise I was severely underpaid 😭. The rsETH<>wETH borrow proposal was launched in one of the most stressful periods of my life, when I witnessed years of my efforts being destroyed in a few weeks, and when the proposal received a green light from both risk teams and tech analysis. I did exactly what Stani and everyone else did: I didn't push back and voted yes.
I have no issue taking my own weight own responsibility in this and will not pretend otherwise.
But I'm not allowing anyone to say my decision was influenced by an angel check.
Show more
$200M was drained from Aave's rsETH markets following an exploit of Kelp's LayerZero bridge earlier today.
The exploiter deposited rsETH, and borrowed 83,427 WETH and wstETH from Aave.
The breakdown of extracted assets across Aave instances is:
Ethereum
52,834 WETH
Arbitrum
29,782 WETH
821 wstETH
At the time of writing:
- Aave rsETH markets are frozen.
- No other LayerZero OFT token was affected.
We are assisting multiple teams in investigating the root cause and the full path the funds took after extraction.
A longer investigative report will follow soon.
Show more
super easy to use; have run this now on all chaos ai pages, and already have a backlog of tasks to work through.
the score is clear, and actionable, awesome job @assaf_elovic and team
This might be the most important score your website will ever get.
Agents are about to become the #1# source of traffic on the internet. Sites that aren't ready will disappear.
spawns real agents on your site and tells you exactly where you stand.
Scan your site in 1 minute →
Show more
🚨 New clip: Imagine a bank with no oversight, no security, and no transparency.
That’s closer to how parts of DeFi operate today than many would like to admit.
@omeragoldberg explains why risk management — not just code — is where things keep breaking.
Watch the full episode 👇🏻
YouTube:
X:
Spotify:
Show more
Chaos holds a simple principle: we only put our name on work we fully believe in.
Principles matter when they cost you something.
Today it's costing us $5 million.
To the Aave community: thank you for the trust. It was a privilege 👻
Show more
Chilling read, but none of this is new.
This playbook isn't unique to the DPRK; it's similar to how every intelligence agency and professional group operates.
The thinking is straightforward:
- What are the most valuable targets? In DeFi, TVL is a great proxy.
- What's the exploit surface?
- What's the probability of successful infiltration?
- Can we launder the funds quickly and make a clean getaway?
Once you've identified which targets are worth hitting and how much you can extract, you determine how much you're willing to invest.
Simple math: TVL is $1B.
The probability of a successful e2e hack is p = 0.10.
EV = 1B * 0.1 = $100M.
If I can execute for less than that, it's worth it.
This may look familiar to Oracle folks, because it's the core of determining the cost of market manipulation...
Back to the Drift exploiters
$1M in capital deposited in Drift + a team of assets interfacing with the org over months; I'd wager their all-in cost basis was under $3M?
That's a no-brainer bet on a $100M+ expected payoff.
Ofc, all of this is oversimplified, but that's the gist.
Most people think hacking looks like the movies; breaking some crazy cipher or a god-tier algo that defeats all your firewalls and circuit breakers.
In reality, humans are typically the weakest link, and social engineering is almost always the easiest way in.
That's why role-based access control, privilege separation, timelocks, monitoring, PagerDuty, and multisig hygiene matter so much.
They're not there to prevent the breach; they're there to limit the blast radius when it does.
Do everything to make sure it doesn't happen.
But assume it might, and design accordingly.
Show more
Chilling read, but none of this is new.
This playbook isn't unique to the DPRK; it's similar to how every intelligence agency and professional group operates.
The thinking is straightforward:
- What are the most valuable targets? In DeFi, TVL is a great proxy.
- What's the exploit surface?
- What's the probability of successful infiltration?
- Can we launder the funds quickly and make a clean getaway?
Once you've identified which targets are worth hitting and how much you can extract, you determine how much you're willing to invest.
Simple math: TVL is $1B.
The probability of a successful e2e hack is p = 0.10.
EV = 1B * 0.1 = $100M.
If I can execute for less than that, it's worth it.
This may look familiar to Oracle folks, because it's the core of determining the cost of market manipulation...
Back to the Drift exploiters
$1M in capital deposited in Drift + a team of assets interfacing with the org over months; I'd wager their all-in cost basis was under $3M?
That's a no-brainer bet on a $100M+ expected payoff.
Ofc, all of this is oversimplified, but that's the gist.
Most people think hacking looks like the movies; breaking some crazy cipher or a god-tier algo that defeats all your firewalls and circuit breakers.
In reality, humans are typically the weakest link, and social engineering is almost always the easiest way in.
That's why role-based access control, privilege separation, timelocks, monitoring, PagerDuty, and multisig hygiene matter so much.
They're not there to prevent the breach; they're there to limit the blast radius when it does.
Do everything to make sure it doesn't happen.
But assume it might, and design accordingly.
Show more
Diving deep on Drift's exploit w/ @laurashin on @Unchained_pod.
This exploit was methodical and calculated.
The exploiters spent time studying Drift deeply.
The game of security/risk is asymmetric:
You only need to be wrong once for it to be over.
Show more
Another week, another DeFi exploit 🫠
@omeragoldberg joined me to unpack the Drift Protocol hack:
⁉️ What went wrong?
👀 How the attack resembles the Mango DAO and Resolv exploits
🤔 Why was Circle so slow to react?
⚠️Are North Korean state actors behind the attack?
Timestamps:
🚀 0:00 Introduction
🥶 0:54 Why the Drift Protocol hack is so chilling
⁉️ 4:32 Was the admin key set up to blame? Or Was it a supply chain attack?
📍 9:17 How the attack is reminiscent of the Mango DAO and Resolv exploits
😬 14:09 How a Solana feature allowed Drift's hackers to lie in wait without triggering alarms
❌️ 19:55 How Drift Protocol failed to implement best practices
🦠24:53 Who else has been impacted by the Drift Protocol exploit?
🤔 27:50 Should Circle have acted faster to freeze the loot?
⚠️ 31:20 Why Omer thinks the Drift Protocol exploit has North Korea written all over it
📝 34:34 Why Omer says the incident calls for better DeFi disclosures and audits
Show more
Working groups and discussions are great, but recent exploits were not novel imo
(I'm not talking about the method in which access was gained, be it phishing, social engineering, etc., but what the attackers were able to do after keys were obtained.)
USR was an unlimited mint on a SERVICE_ROLE.
Drift was a 2/5 multisig with 4 new signers and 0 timelock.
So, I'd say the two aren't mutually exclusive.
You need coordination AND you need to actually give a care about opsec, access controls, and privilege separation.
Right now, too many teams treat risk/security as someone else's problem until it's everyone's problem.
Crypto also has a penchant for reinventing the wheel. We love to rename things! But solutions for different threats already exist
- CrowdStrike
- Palo Alto
- Wiz
- etc
We don't need to rebuild Web2 security from first principles. We need to adopt what works and focus energy on the stuff that's actually novel, which actually sounds like what @andrewhong5297 is describing here.
AI and better tooling can make the technical stuff less intimidating, but only if there's a culture/standard around it
Show more
Kudos to the team on posting a transparent post-mortem.
Supply chain attacks are skyrocketing; you should do everything to ensure it doesn't happen, but operate under the assumption that it can.
Developers install 100's of libraries, which in turn can have thousands of dependencies. It only takes one compromised lib to potentially get root on your machine or access CI/CD secrets.
With this in mind, any asset issuer should do a threat model:
what are the attack vectors?
what's the blast radius for each?
The most painful part in reading it is that it could have been prevented with onchain sanity checks and guardrails that take an afternoon to implement.
For any asset issuers:
- use a multisig for sensitive operations, especially mint/burn
- add timelocks to those operations
- add velocity-based sanity checks - i.e., a 40M USD mktcap token should not be able to mint 100m in a minute/hour/day; even if this adds operational friction!
- use a proof of reserve oracle - these are relatively easy to implement and cheap
- add risk oracles as circuit breakers
For offchain opsec and cyber use, CrowdStrike / Palo Alto Networks, etc. These aren't bulletproof either, but this class of attacks is common, and many of these signatures and malwares have identifiable signatures that can potentially be blocked/caught.
Monitoring is important, and everyone should have it, but the best actions are preventative.
Good luck to the Resolv team and community with the continued investigation 🙏
Show more
This is why we can't have nice things
How Tradfi Values Crypto With Oil at $100
1/ Drift's admin key was compromised.
$213M+ drained from @solana's largest DEX in under 10 seconds.
Unfortunately, we've seen similar patterns before:
- fake collateral market
- a manipulated oracle
- disabled circuit breakers
Let's break it down 👇
written w/ Chaos AI
Show more
