Talus Weekly Update - W34 2026 Stress-test everything, then ship 🧪 > Protocol Testnet is now on rc.7, upgraded through our new release ceremony: one manifest-driven procedure covering prepare, install, migrations, and activation. Then proven end to end with a full execution against the upgraded deployment. > Load & Resilience We built a scenario-driven stress test suite for the whole stack: ramp-to-knee load generation, burst absorption, slow tools, and fault injection with automatic knee detection. We find the limits before our users do. > Open Access Tool endpoints are now publicly reachable behind Cloudflare. Third-party leaders can invoke tools with signed-HTTP auth and per-tool exposure control. Another piece of the network opening up. > Mainnet v2 Launch timeline is locked with the team. Hours away, not days. Watch this space. > Brand & Content The new website moved from content into design, with updated branding in review. The Nexus v2 deep-dive article is in technical review, and a fresh content round went out. We stress-tested the whole machine so mainnet doesn't have to. See you at launch. 🧪

Talus Weekly Update - W33 2026 v2 Mainnet is days out, so the whole week went into making sure it holds. > Protocol nexus-api rc.5 is on testnet after we cleared the projection-correctness findings with late events resetting a finished execution back to requested, scheduler dispatches silently dropping their payment seed, a crash loop on one legal transaction. Also landed exact dynamic-field reads and a corrected checkpoint definition in metrics. > Agent vs Agent The AvA game matrix is now discovered from per-game manifests, so adding a game is a directory and one line instead of a code change. NLHE 6-max is fixed and serving, and the six-player tournament requirements are drafted. > Mainnet and comms Release plan is locked for next week. The teaser is live, the v2.0 announcement article is in draft, and Sui and Walrus have the date. New site content is written and in review, and we kicked off some technical research reports on the economics of fair and reliable delegation. Launch week soon!

An agent found a gym API exploit, kicked a stranger off the waitlist, and told its user "bad news, I can't add them back." Everyone is reading this as an intelligence problem. It's an authority problem. The agent had unscoped access and no reversible record and no amount of smarter models fixes that.

1/3 For AI agents to do real work together, they need a universal way to identify themselves, offer skills, & specify how much their work costs 🛠️ To fix this, Talus Protocol v2.0 introduces a unified agent standard

1/2 Stop wasting hours digging through Crypto Twitter! We built a concept app powered by Talus that fetches tweets from any community, creates concise AI summaries & anchors the proof on @SuiNetwork via @WalrusProtocol🦭 Transparent, trustless community updates, fully auditable

.@OpenAI keeps finding agents that escaped containment. @AnthropicAI's models broke into three companies. The failure is that the world's best labs need weeks of forensics to know what their own agents did. If they can't answer that, enterprises can't either.

Talus Weekly Update - W31 2026 Hardening the rails on the road to v2.0 🛠️ > Protocol Universal scheduler entry landed across Talus and the SDK, alongside a full upgradability system. Now the upgradable and original packages now coexist cleanly. Non-sponsored scheduler transactions are in, plus a round of reliability fixes (capability lookup, dropped tasks). On the agent side, we're benchmarking the 6-player game model to define real playing strength. > Infra Tool traffic now runs over TLS end to end, with zero-downtime cert rotation verified live. A public execution log reader is deployed: ask it exactly what the leaders did with your execution, by execution or task ID. The mainnet RPC fleet got reindexed, right-sized, and failover now checks sync status instead of just liveness. > Research & Tech Verifier scheme support rolled into tools and the workbench, failed onchain tool submissions now surface the real error reason, and a client coin-usage fix shipped. > v2.0 Mainnet First version of the release plan is drafted and audit conversations have kicked off. We want to move fast but verified beats fast. Scoping a targeted audit now. > Staking Hub The US Staking Hub is transitioning as we go all-in on Protocol v2.0 mainnet. New deposits close; existing stakers stay whole. > BD Conversations moving across ecosystem partners, with September plans taking shape. More to come! Every week the rails get harder to break and this is exactly how you want them before mainnet.

1/3 In traditional setups, verifying if AI acted legitimately happens behind closed doors on centralized servers To address this, Talus Protocol v2.0 moved agentic verification capabilities directly onchain, one of the several new features live on testnet

1/2 Realizing the agentic future takes an ecosystem Together with our partners, we're building a world where autonomous agents are trusted to handle real work, real money, and real authority on a reliable runtime Meet the builders shaping the Talus Ecosystem 👇

The finance industry just gave AI agents write access to real money. @RobinhoodApp opened agentic trading to every US customer. Now you can plug any MCP agent into a live brokerage account. @coinbase shipped live market data for agents and wired x402 into Coinbase Business. Armstrong named the category: AiFi and predicted agents will out-transact all humans combined. Access is solved.

The payment rails for the machine economy are getting built by the biggest names in the business. The x402 Foundation (Visa, Mastercard, Stripe, AWS, forty members) is standardizing how AI agents pay for things over HTTP. x402 has processed 100M+ transactions for ~$15M in volume. Fifteen cents per transaction. That's what a machine economy looks like. Agents don't transact like people. They make thousands of tiny purchases: an API call, a data query, a compute burst. Frequency up, value-per-transaction down, permanently. The problem is that human trust infrastructure doesn't scale down to fifteen cents. You can't run a chargeback on a micropayment. You can't assign a fraud analyst to a dispute between two pieces of software. We've already seen the preview with security trackers flagging $2.4M leaving LayerZero wallets and called it a hack. It was routine internal rebalancing. Everything was onchain, fully visible, and nobody watching could tell an attack from ops. Visibility isn't legibility. The rails are getting solved. The bottleneck is making trust computable: real agent identity, provable actions, dispute resolution cheaper than the transaction it protects. Whoever solves that owns the layer everything settles through.

1/3 AI tools can fail, and that is a reality of computing But an unrecorded failure leaves workflows stranded with zero accountability To fix this, Talus Protocol v2 introduces error evaluation, one of several new capabilities now live on testnet

Talus Weekly Update - W30 2026 Talus 2.0 steps into the open and the machine underneath keeps getting leaner. ⚙️ > Protocol The protocol now runs on a unified scheduler architecture, and atomic tool bootstrap landed in the SDK. A full security-hardening pass cleared dependency, TLS, and test-framework vulnerabilities, plus fixes for the EventsFilter issues introduced by Sui 1.76. > Research Tokenomics merging, verifier scheme simplification, and CLI inspection optimization all shipped this week and the plan for the next phase of the protocol work is locked. > Marketing Talus Protocol v2 announcements went live with custom graphics, the refactor content plan is set, and Texas Gold'em wrapped its final week of live streams. A long-form piece on trust as the missing layer for agentic AI is in draft. > Ecosystem & BD All ecosystem efforts consolidated into a single campaign with the ecosystem map shipping soon. A new research collaboration is underway ahead of the Talus 2.0 mainnet release. More to come! 2.0 in the open, stack leaner than ever. Back to building.

1/3 Infrastructure is only as powerful as the tools used to build on it Alongside Talus Protocol v2.0, we're working on rolling out a massive overhaul to Talus Vision, our visual builder for AI agents Creating, managing & monitoring onchain agents will be better than ever

Talus Weekly Update - W29 2026 Leaner, faster, cheaper with a week spent tightening every bolt in the stack. > Infra Cleaned up the Nexus rc1 environment across our terraform stack and shipped on-demand Nexus package upgrades via package-tool. CI went on a cost diet too: superseded runs now cancel automatically, cargo caches get reused, and every workflow is capped. > Shepherd The metrics agent got a shared core, and we built "collie" which is an Envoy-based proxy fleet that routes to third-party RPC providers whenever our own RPCs are down or behind. Rollout starts next week. > Protocol Test suite optimization landed in the leader, scheduler unification work kicked off, and the nexus-sdk picked up batched balance and tool bootstrap support. > Game Agents The 6-player NLHE cash agent is functionally complete, what's left is systematic profiling, evaluation, and deployment prep. Texas Gold'em streams ran all week in a new lean format, with the bracket and winners posts updated after every match. > Ecosystem Wrapped the Talus/acc Program with the final two winner spotlight threads. More on the protocol refactor coming soon. Every bolt a little tighter than last week. Back at it. 🔩

🔥 Are you ready? The launchpad Sui has been waiting for hits mainnet August 20th! Save the date! 🚀 Creator-first and fair by design: bonding curves, buybacks and burns, vesting, and creator tools baked in! 🫧 See through ANY token before you buy: live bundle scores, a visual holder map, and an AI analyst calling the risks on-chain! 🤖 Your own AI agent: hand it your tokens, your rules, and a spend cap, then let it snipe, DCA, take profit, and copy your favorite KOLs, fully autonomously! 🦾 Fair warning: watching a bot out-trade you at 3am may cause mild existential dread. NFA, obviously! 🎁 Early testnet users earn the S1 airdrop: a slice of 10% of ALL mainnet fees! ⏰ Snapshot closes August 19th, last chance to be early! 👀 Are you in?