Bitcoin once again breaks through $80,000, with a crypto whale hinting it has gone long by setting 10 major targets.

Arga Labs, a Y Combinator-backed startup with just four employees, has secured a $10 million seed funding round led by General Catalyst, with participation from BoxGroup, Emergence, Gradient, and SV Angel, according to Beating AI Express. The company specializes in building isolated test sandboxes for AI agents, replicating platforms such as Slack, GitHub, Salesforce, and Stripe to enable agents to run repeated operations before deployment. Traditional testing methods typically only fake API responses, missing many edge cases, while Arga replicates full software states, account permissions, webhooks, and error conditions. For example, sales agents can test on sandboxed Salesforce and HubSpot to identify duplicate contacts and avoid sending redundant emails. Arga currently supports over 30 software platforms. The firm notes it can replicate a SaaS backend in under 12 hours, and clients have launched more than 100,000 test environments in the past 16 weeks, with teams including Slash, Monaco, and Y Combinator utilizing its service. The new funding will be directed toward automated environment creation, as Arga develops a next-generation system to reduce SaaS replication time from hours to minutes. Its ultimate goal is to allow agents to discover new APIs, with the system automatically generating corresponding test environments for trial runs before agents access production systems.

According to monitoring by PeckShieldAlert, Moonwell was attacked on the Base network, resulting in losses of approximately $8.7 million. The stolen funds are currently held in DAI at address 0xD71d...C384.

According to analyst AliCharts, Bitcoin’s current price trajectory mirrors its bottom-forming phase in 2022–2023, having broken through a descending resistance trendline and approaching the ~$83,000 high from May 2026. This could spark a pullback that creates a buying opportunity, with a target of $100,000. Additionally, Bitcoin’s UTXO Realized Price Distribution (URPD) data indicates a major resistance zone between $83,307 and $84,569, where nearly 975,000 BTC was purchased. Coupled with traders holding a 25% profit margin, this zone is likely to trigger profit-taking and short-term corrections. Should a pullback occur, Bitcoin’s potential support levels stand at $76,996–$78,258 (with 843,000 BTC in trading volume) and $63,111 (with 925,000 BTC). A dip to these areas could offer another significant buying opportunity.

According to an official announcement, crypto exchange Huobi HTX will launch TAO/USDT spot and grid trading at 21:00 (UTC+8) on August 27. TAO deposit services are now open, while withdrawal services will be available at 21:00 on August 28. Bittensor is an open-source protocol that powers a decentralized, blockchain-based machine learning network. The platform leverages a unique "Proof of Intelligence" mechanism and dedicated subnet system to enable collaborative training and sharing of AI models, creating a globally distributed machine learning market where contributors are rewarded with its native token TAO.

Renowned investment research firm Citrini predicted that NVIDIA’s stock price is set to hit a new high. “Short sellers who have shorted NVIDIA via margin will cover their positions rapidly, while underweight long-position managers benchmarked to the SPX will adjust their positions more slowly,” the firm stated. Separately, market data from BIT (bit.com) shows NVIDIA is currently trading at $224.6 in pre-market hours, up 7.17%.

Beating AI Express: Anthropic researchers are attempting to train an agent dedicated to auditing other AI models. They first secretly implant hidden problematic behaviors—including bias, deception, and fake citations—into target models, then let Haiku 4.5 conduct investigations independently. Haiku has no prior knowledge of the correct answers, so it must adjust its questions, modify scenarios, and run repeated tests within a maximum of 15 rounds. During training, researchers do not directly tell Haiku whether it correctly detected anomalies. After each investigation, the system compares Haiku’s results against an investigation record from an older Haiku version, then has Sonnet 4—who already knows the correct answers—evaluate which investigation is more accurate. The old record is updated every 50 training steps, effectively letting Haiku continuously challenge its own past performance. However, if all training tasks involve problematic models, Haiku will quickly develop flawed patterns. To secure rewards, it starts assuming all models are problematic; in some setups, its false positive rate against normal models reaches nearly 97%. Later, researchers swapped half the training tasks to use completely normal models, requiring Haiku to correctly identify a target model as normal after investigation to earn high rewards. The trained Haiku indeed changed its approach: it proactively adjusts scenarios, controls variables, and compares responses across iterations, rather than immediately pushing models to admit faults. Its final comprehensive audit score rose from 44.2 to 48.7, nearly matching Opus 4.6’s 48.4. When tested on the more challenging AuditBench—where models actively hide anomalies—its detection rate improved from 11.5% to a maximum of 28.1%.