Bitcoin spent Friday getting knocked around. Price rejected hard from above $81,000, dumped below $77,000 after Fed Chair Kevin Warsh's Jackson Hole speech, and triggered nearly $488 million in liquidations. Spot ETFs snapped a 9-day inflow streak with $201 million in outflows. Meanwhile, xAI put Grok 4.6 on both Microsoft Foundry and Amazon Bedrock GovCloud, aiming straight at enterprise wallets. And on the quantum front, BIP 360 just got merged into the Bitcoin Improvement Proposals repo, while Blockstream published SHRINCS, a hash-based post-quantum signature scheme. Busy weekend. Let's get into it.
Bitcoin is not having a good time. On August 28, it poked above $81,000, hit an intraday high around $81,455, and then got smacked back down to the mid-$76,000s. As of the weekend, it's trading around $78,000, sitting uncomfortably between $77,000 support and $80,000 resistance.
The trigger was Fed Chair Kevin Warsh at Jackson Hole. Markets were hoping for dovish signals given softer inflation prints. Warsh went the other way, dismissing recent low inflation readings and reviving the possibility of higher rates. Crypto responded with a $488 million liquidation cascade. Spot Bitcoin ETFs ended a 9-day inflow streak with $201.8 million in net outflows on Friday, led by ARK 21Shares, and total ETF assets slipped back below $100 billion.
Now the structural picture. Glassnode is flagging a dense liquidity zone between $81,000 and $86,000, packed with long-term holder supply that hasn't moved in at least six months. That's a wall. Above that, the 365-day VWAP sits near $82,600. Below, the 50-week and 100-week EMAs are around $77,300 and $78,500, right where we're trading now.
Order book data from HTX shows about $5.7 billion in bids clustered between $75,000 and $78,500, versus only $2.8 billion in asks above $81,500. That skew tells you where the pain trade lives, and it's lower before it's higher.
One slightly more optimistic signal: a pseudonymous analyst flagged that Bitcoin's on-chain profit-loss crossover is reappearing. Historically that pattern showed up near cycle bottoms in 2015 and 2019. Take it with salt. It's a lagging indicator, and this cycle has more institutional plumbing than any prior one.
And about those whale accumulation headlines. Yes, large holders added roughly 43,000 to 46,000 BTC over the last 60 days. But a chunk of what trackers label as whale buying is actually ETF custodial wallets absorbing inflows. The same "whales are buying the dip" headline ran at $92,000 in January. Read the wallet labels before you read the narrative.
xAI just made its most serious enterprise play yet. Grok 4.6 is now live on Microsoft Foundry and on Amazon Bedrock in AWS GovCloud. Two of the three biggest enterprise AI distribution channels in the world, in the same news cycle.
Grok 4.6 comes with a 500k token context window and four configurable reasoning levels: low, medium, high, and xhigh. It's positioned as a frontier model for long-running agents, coding, and complex visual and interactive tasks. On Bedrock GovCloud, it runs on the standard bedrock-runtime endpoint with Responses, Chat Completions, and Converse APIs, plus cross-region inference routing. On Foundry, you get managed endpoints and enterprise governance controls baked in.
The GovCloud piece matters more than it sounds. GovCloud is where US federal agencies and defense-adjacent contractors run workloads. Getting a model certified there is a slow, expensive process, and it puts Grok in a market where OpenAI and Anthropic have been building for years. For a model that a lot of people still associate with X posts and Elon's Twitter personality, this is a real repositioning.
Separately, xAI is expanding Grok Bot access. Grok Bot is their persistent workflow automation agent, running on its own cloud VM, able to sign into apps, manage files, join meetings, take notes, process refunds, even deploy websites. It's now bundled into SuperGrok Plus, Cursor Pro Plus, and Cursor Teams plans. And they're already teasing Grok 5 by year-end.
What this looks like from a distance: xAI is trying to compress two years of enterprise sales motion into a few months by riding Microsoft and Amazon's distribution. Whether Grok 4.6 actually beats GPT-5 or Claude on real workloads is a different question, and Foundry's whole pitch is that you can benchmark them side by side. Enterprises are about to find out.
Healthcare AI just crossed a line that's been talked about for a decade. The FDA has now formally accepted that autonomous AI diagnostic systems can operate without a physician reviewing the output first. LumineticsCore, formerly known as IDx-DR, is the reference case: it screens for diabetic retinopathy and delivers a clinical recommendation in primary care, no ophthalmologist required. Aidoc got 2026 clearance for a single triage AI that flags multiple urgent conditions across hospital imaging workflows. Paige Prostate Detect is doing similar things in pathology.
At the deployment layer, things are moving fast. FDB announced the first commercial rollout of Script Agent, an AI prescribing agent that listens to the patient visit conversation and turns it into structured prescriptions for the clinician to approve. It's launching inside Tebra's platform for independent practices. Grounded in FDB's medication database, not just raw language models, so it flags overrides when prescribing falls outside normal practice.
Sheba Medical Center in Israel is building what they're calling an AI-powered hospital, starting in the emergency department. They're OpenAI's first international hospital partner. Their SmartER system automates patient encounter documentation and gives clinicians live access to medical literature through ChatGPT for Healthcare.
Then there's DeepCura's Electronic Health Workforce, or EHW. Their pitch is that the EHR is dead as the center of gravity. Instead, you get a supervised AI workforce, receptionists, scribes, intake nurses, all producing a receipt for every action. They claim 6,000 clinicians already using it. And Epic launched Agent Factory at HIMSS26, letting hospitals build no-code AI agents directly inside Epic's EHR, powered by Epic Cosmos data covering hundreds of millions of patient encounters.
The pattern here is clear. AI in healthcare is moving from behind-the-scenes assistant to named participant in the workflow, with audit trails and clinician sign-off. The unresolved questions are liability, real-world evidence, and what happens when the AI is wrong. Autonomous prescribing is still not FDA-cleared. But everything short of that is being productized right now.
Bitcoin developers are quietly making real progress on post-quantum defense, and this week produced three separate data points worth stitching together.
First, BIP 360 was merged into the Bitcoin Improvement Proposals repository. Merged, not activated, but that's the formal starting gun for review. BIP 360 introduces a new Pay-to-Merkle-Root output type, disables key-path spending to protect public keys when coins move, and lays groundwork for adding post-quantum signature schemes in a future soft fork.
Second, Blockstream published a BIP for SHRINCS, a hash-based, stateful post-quantum signature scheme. Stateful means your hardware wallet has to track which keys it's used, because reuse breaks the security. Signatures grow with each use, and losing a device requires a fallback transaction around 5,777 bytes. It's promising but explicitly not cryptographically mature yet. A full security proof is still outstanding.
Third, StarkWare pulled off the first on-chain test of a quantum-resistant Bitcoin spend using their Quantum Safe Bitcoin scheme. Block 964,199, a 10,000 satoshi output spent, processed via MARA's Slipstream service because the transaction is nonstandard under default relay policies. Cost to generate: roughly $150 to $200, plus hours of computation. The point wasn't cheapness, the point was proving Bitcoin's existing consensus rules already allow quantum-resistant spending without a hard fork.
On urgency, the experts don't agree. Caltech's Thomas Rosenbaum thinks fault-tolerant quantum systems could arrive in 5 to 7 years. Jameson Lopp at Casa thinks it's decades away. NIST sits in the middle. One data point worth noting: an Anthropic model reportedly cut the work needed to break a leading post-quantum signature candidate by a factor of 67 million last month. That's the kind of thing that shortens timelines whether you like it or not.
And zooming out from quantum specifically, the covenants conversation is heating up too. Cofund published an interactive atlas mapping 24-plus Bitcoin covenant use cases across CTV, OP_CAT, and CSFS proposals, from vaults to payment pools to coordination-free mining pools. Not everything is coded, but the design space is being mapped seriously. Bitcoin's upgrade process is slow. That's the feature. But there's more activity in the pipeline than the price action would suggest.
One prediction: by the time Bitcoin actually needs quantum-resistant signatures, half the exchanges and custodians using the phrase "quantum-safe" in their marketing today will still be running the same elliptic curve keys they have now. Consensus is slow. Marketing is fast. Watch the code, not the press releases.