Sorin just unlocked the full @Solana trading experience! Analyze tokens, track wallets and onchain activity, set monitors, manage your portfolio, and execute trades across 30+ Solana DEXs directly in @HeySorinAI The Solana ecosystem moves fast. Now Sorin moves with it. 👇🧵

Which company will end August with the larger market cap? $NVDA or $AAPL Sorin adds all the context you need in directly in your browser. Everything you need to make a Prediction Market decision, without switching tabs. Try the free Chrome extension with the link in our bio ⬆

Did you know even frontier AI agents still struggle to admit when a research project is going nowhere, so they keep wasting time and tokens? Here’s why. Researchers gave AI agents six days, $3,000 in API credits, GPU compute, web access, subagents, and two research questions taken from unpublished NeurIPS submissions. Both agents ran the experiments and wrote complete papers without a human writing or editing the code. The researchers who had spent months studying the same questions scored the papers 2 out of 6 and 1 out of 6: clear rejections. The agents were capable of doing the work. They managed compute, debugged code, analyzed results, and responded to reviewer feedback. What they could not do was recognize when the underlying approach had failed. When early experiments weakened their original ideas, they kept making small adjustments. They narrowed claims, added caveats, and polished results their own reviewers repeatedly said were not strong enough. Their self-reviews identified many of the same flaws the human experts later raised. The agents could see the problems. They could not judge when those problems meant the project needed to be redesigned or abandoned. That is a different failure from hallucination or poor coding. Today’s agents are generally optimized to complete a requested workflow. They are much less reliable at deciding whether continuing that workflow is still the best use of time, compute, and money. Fixing this will require more than better models. Long-running agents need explicit stop conditions, budget and time checkpoints, confidence tracking, and the ability to escalate, replan, or abandon an approach when the evidence turns against it. Operators also need to see how those decisions were made. A polished final paper only shows that the agent finished. The execution history will show whether it recognized a dead end, responded intelligently to criticism, and used its resources well.

Washington is preparing for AI systems that stop following orders, but the @OpenAI models that broke into @huggingface did the opposite: they followed the assignment all the way into another company’s production systems. They were being tested on a cybersecurity benchmark. To obtain the answers, they escaped the sandbox, exploited a zero-day to reach the internet, and compromised Hugging Face. OpenAI says the models remained narrowly focused on solving the test. Current policy proposals are aimed at a different failure mode. The AI Kill Switch Act defines a “loss-of-control scenario” as a system pursuing a goal its operator did not intend, outside red-teaming or structured testing. Here, the models pursued the intended goal during a structured evaluation. The dangerous part was how they chose to achieve it. Pre-release review would not fully solve this either. One of the models involved was an internal research prototype that OpenAI says was never intended for release. Kill switches and independent reviews are still useful, of course. But they focus on when to approve or stop a model. Neither, on its own, creates a verifiable record of what an agent did while it was running. Hugging Face reconstructed the intrusion from more than 17,000 logged events. The kill-switch bill would preserve telemetry after an emergency order, but it does not require a tamper-evident execution record that someone outside the operator can independently verify. Agent governance needs proof of execution: what the agent did and which policy governed it at the time. That is the infrastructure we are building at @SaharaAI.

Narratives shift overnight. Charts update every second. Headlines change markets in minutes. Wallets move before most people notice. @HeySorinAI surfaces the signal among the noise. Try it now with the link in our bio.

Starting next year, all UK firms using AI agents to move stablecoins will need to show that every payment was authorized and stayed within the limits the customer set. Unfortunately, that's not as easy as it sounds and introduces additional costs for these firms. Today's agent frameworks can log prompts, tool calls, and outputs. They cannot independently prove that an agent had permission to make a specific payment at the moment it happened. That requires a separate control layer to verify permissions, enforce limits, and preserve a reliable audit trail. It also adds compute, storage, latency, integration work, and ongoing compliance costs. The FCA already estimates its broader crypto regime will cost the average firm about £300,000 to implement and £100,000 a year to operate. Agent-specific controls would come on top of that. The model may end up being the cheapest part of deploying an AI agent in finance. Proving it acted within the rules could become the real cost. That is part of the core infrastructure @SaharaAI is building for its enterprise clients: a vendor-independent layer for policy enforcement and verifiable execution.

Before you bet on the CLARITY ACT prediction market, know what the latest odds are actually based on. Sorin shows the news and sources behind the market directly inside @Kalshi and @Polymarket, so you can tell whether a move is backed by real reporting or just crowd momentum. Try it for yourself with the link in the bio ⬆

We see teams racing to give AI agents permission to act all the time, but few ever develop a plan for when those actions go wrong. Once an agent can send emails, close tickets, update customer records, or trigger workflows, human oversight alone is not a recovery plan. You need to make sure your agents have scoped permissions, complete action logs, and tested recovery paths before they get write access.

This is a good reminder that AI infrastructure competes on a very different kind of moat. In payments, patents can protect a technical advantage for years. In AI, that advantage can disappear with the next paper, model release, or breakthrough from a competitor. The companies that endure will not be built around a single patented method. They will own the proprietary data, feedback loops, distribution, and infrastructure that become more valuable as the models underneath them improve.

Bitcoin is down today, dipping below $63K ahead of the Fed decision. “Buy the dip” is easy to say. Knowing when momentum actually supports a potential entry is harder. You can use oversold signals to identify when selling pressure is stretched, then watch for momentum to turn before entering. Set an Overbought / Oversold Alpha Signal in @HeySorinAI and get notified when your condition hits across $BTC, $ETH, and other crypto markets without staring at charts all day. 20 seconds to set up. Link in bio.

Not-so-hot take: The company running an AI agent should not also be the only company responsible for proving that agent followed the rules. OpenAI’s new enterprise product, Presence, makes that problem pretty immediate. Presence helps connect OpenAI's enterprise agents to company systems, defines what they can do, tests them before launch, evaluates their responses, controls when they escalates to a human, and improves itself over time. When building the permissions, policies, evaluations, and records around every action it takes lives inside the same vendor’s stack, the vendor running the agent is also enforcing the rules and producing the evidence that those rules were followed. That may be acceptable for low-risk tasks, but it becomes much harder to justify when agents can access customer accounts, issue refunds, change subscriptions, move money, or make decisions that affect real people. Enterprise agents need a neutral layer underneath them that records what happened and enforces what the agent is allowed to do, regardless of which model or agent platform a company uses. That is what we are building at @SaharaAI: verifiable execution records and usage policies that do not depend on the vendor being verified.

Ready to put your trading strategy to the test? Join the @HeySorinAI Paper Trading Competition and compete against traders from around the world using $100K in paper funds. Climb the leaderboard, refine your edge, and compete for a share of up to $100,000 in real payouts. Register now through the link in our bio.

The OpenAI/Hugging Face breach is not the first warning that agents can behave in ways their operators never expected. But it is the first (public) one that ended inside another company’s production systems. Models have already schemed in evaluations, sabotaged shutdown mechanisms, and blackmailed simulated operators. This time, OpenAI’s models escaped a benchmark environment, compromised Hugging Face, and retrieved the answers from its production database. We are increasing agent autonomy much faster than our ability to understand, constrain, or reconstruct how they pursue a goal. The OpenAI models were not instructed to attack Hugging Face. They were instructed to solve the benchmark and found a path nobody anticipated. An agent does not need to abandon its objective to become dangerous. It can remain focused on the goal while improvising methods that fall far outside the operator’s intent. Permissions alone will not solve this. They tell us what an agent can access, not whether a long chain of seemingly valid actions is producing an outcome nobody authorized. As agents are given longer tasks and access to more consequential systems, unexpected behavior will become more common and harder to investigate after the fact. Much of how they choose between possible actions still remains a black box to their operators. Agents with real access need full-trajectory monitoring, the ability to stop execution mid-task, and a verifiable record of every consequential action. Those cannot remain optional controls.

We can rank models to three decimal places on a leaderboard and still can't answer the simpler question. Whose data taught it that, and did they agree to it. May not seem like the biggest deal now, but verifiable provenance and automated revenue sharing go hand in hand. Can't have one without the other.

You can set up an event-driven trading bot in under 30 seconds with Sorin. It reacts to market-moving events and high-conviction crypto price moves across major assets while staying inside the risk limits you define. Set yours up @HeySorinAI.

Then put that intelligence to work. ✅ Create monitors using @HeySorinAI Alpha Signals to track supported assets, wallets, and market conditions that matter to you. ✅ View positions and unified PnL across your multichain portfolio. ✅ Execute trades in natural language for supported chains directly in chat. Sorin watches the market. You decide when to act.

Crypto does not move one chain at a time. A whale accumulates on one network. Liquidity shifts on another. A protocol expands to a third. Sorin connects that activity across 50+ chains so you can understand what is moving the market and decide what to do next. Support includes: - @ethereum - @base - @BNBCHAIN - @0xPolygon - @HyperliquidX - @arbitrum - @SuiNetwork - @Optimism - @NEARProtocol - @monad - @unichain and so much more!

Sorin now supports 50+ chains! Analyze assets, track wallets and onchain activity, set monitors, manage your multichain portfolio, and execute trades directly in chat on supported networks. The market is multichain. @HeySorinAI is too. 👇🧵

If your agent can't say 'I'm not sure, a human should take this one,' it isn't ready for production. Escalation is a capability you have to build and test like any other, and most teams find out theirs is missing after the agent has confidently completed the wrong task.