Exchange Ranking
Top Exchanges
24h Volume Ranking
Popularity Ranking
Exchange BTC Balance
Proof of Reserves
Decentralized Exchanges
Funding Rate
Funding Heatmap
Liquidation Data
Max Pain
Long/Short Ratio
Whale L/S Ratio
Binance/Okex/Huobi L/S
Bitfinex Margin L/S
ETF Tracker
Solana ETF
XRP ETF
Hong Kong ETF
Bitcoin Treasuries
Crypto Reversal
Ethereum Reserves
HyperLiquid Wallet Analysis
Hyperliquid Whale Watch
Large Transactions
On-chain Movement
Bitcoin ROI
Stablecoin Market Cap
Options Analysis
News
Articles
Economic Calendar
Features
Wallet
Contract Calculator
Security
Collections
Watchlist
FollowingKey Points
OpenAI said on August 7 that the possibility that Astra has critical cybersecurity capabilities cannot be ruled out.
The company suspended some internal Astra work while strengthening security controls, sandboxes and surveillance.
Astra also produced 10 research results that solved or advanced long-standing problems in mathematics and theoretical computer science.
OpenAI's cybersecurity risks
OpenAI said on August 7 that recent internal assessments have shown significant progress in proxy coding and cybersecurity enough that it no longer rules out the possibility of Astra reaching critical levels under its preparation framework. The company does not claim that the model has reached that threshold and its evaluation is still in its preliminary stages. According to OpenAI's framework, critical models can develop autonomous functional zero-day vulnerabilities that target many critical systems, or perform novel end-to-end attacks from high-level targets. OpenAI said it is tightening network and tool access, model weight protection, encryption and sandboxing, while expanding its monitoring of the use of Astra agents. The company also suspended Astra activities that had not yet met requirements for stronger controls and said it would work with government agencies and selected AI security organizations to conduct testing. OpenAI described Astra as "our next major model" on August 1 and called it "one of our upcoming models" on August 7.
Mathematical debate
Before the cybersecurity warning was released, OpenAI disclosed that the internal version of Astra had produced 10 results in areas such as quantum complexity, lattice cryptography, coding theory, and extremal combinatorics. OpenAI said the work solved or significantly advanced long-standing openness issues and cost approximately $2000 (based on Sol API rates) in tokens. Andrew Blumberg, a professor at Colombia University and a member of the First Proof board, told the media that the results are impressive, but have not changed his view of current AI capabilities. He said many of the results were concise counterexamples or "tiny, extremely clever constructs" based on known work. "They didn't change my transcendental beliefs." These results still demonstrate how cutting-edge models can be useful research collaborators, while also raising parallel security issues when the same coding capabilities are available for software vulnerability exploitation. This tension is now shaping the way OpenAI evaluates access and deployments. Prior to Astra, OpenAI evaluated GPT-5.6 Sol as a high network security level rather than a critical level. An earlier model could assist in vulnerability discovery and exploitation tasks, but OpenAI said it has not yet shown reliable autonomous attacks on hardened targets, making Astra's latest assessment a significant upgrade.
