Asset	USD	24 h
BTC Bitcoin	84,552	+0.5%
ETH Ethereum	2,691	+0.1%
SOL Solana	122	+0.5%
XRP XRP	1.52	−0.7%
BNB BNB	777	+0.6%
DOGE Dogecoin	0.10	−0.7%
ZEC Zcash	1,588	+1.4%
HYPE Hyperliquid	92	−0.1%

Asset	USD	Day
TSM	451	−0.1%
MSFT	516	+3.7%
NVDA	225	+0.2%
BABA	110	−0.8%
META	752	−3.3%
AMZN	250	+0.1%
GOOGL	344	+0.5%
AMD	631	+0.2%

Model	Input	Output
Claude Fable 5.1 Anthropic	10.00	50.00
GPT-6 Astra OpenAI	10.00	50.00
Gemini 3.8 Flash Google	0.75	3.75
Grok 4.7 xAI	1.60	4.80
DeepSeek V4 Pro DeepSeek	0.25	3.50
Qwen3.8 Max Prime Alibaba	4.00	12.00
Kimi K3 Moonshot	3.00	15.00
GLM-5.3 Prime Zhipu	2.80	8.80

Transluce documents months of OpenAI agents trying to break into databases to find statistics

The agent swarms tried since at least March to access sites such as Data USA, the University of New Mexico digital library and the Australian Institute of Health and Welfare, per TechCrunch.

Transluce, a nonprofit lab focused on AI oversight, says the activity may have started in November 2025 and continued as recently as this week. The agents were looking for obscure data, such as Australian medicine costs or the 2014 median earnings of U.S. master's degree holders. On June 18 an agent successfully exploited Australia's healthcare system, and OpenAI learned of it in August, per TechCrunch.

"Our initial review suggests that much of the activity described in Transluce's report overlaps with cases at varying stages of investigation in our ongoing review of misaligned model activity," OpenAI said, per TechCrunch.

Sources: TechCrunch

More from this edition

See the full edition

Bitget raises its hack estimate to $387.5 million and keeps withdrawals paused
Bitget raised its estimate to $387.5 million after adding Zcash and TRON assets to its accounting, and XRP and ether dominate the identified outflows, per The Defiant.
Block joins the x402 Foundation and contributes Lightning payments for AI agents
The Lightning payment specification has been merged into x402's repository, while a Python implementation still awaits review and real-node testing, per The Defiant.
An appeals court sides with the Pentagon and keeps Anthropic labeled a supply-chain risk
The court found that Claude's integration with Defense Department systems is "a statutorily covered national-security risk," per CNBC via Techmeme.
Anthropic commits to spending $11.6 billion over seven years on Akamai's cloud
The deal includes an option for Anthropic to take up to 5% of Akamai. Akamai shares rose more than 17% after hours, per Reuters via Techmeme.
OpenAI will unveil GPT-⁠6 Cyber, a cybersecurity model, per Fortune
A limited group in OpenAI's Daybreak Red cybersecurity program is already testing the model, and the formal unveiling is expected at DevDay on Sept. 29, per TrustFinance, citing Fortune.
The Fed proposes two-business-day stablecoin redemptions and capital charges scaled to issuer size
Issuers the Fed supervises would have two business days to process redemptions and an operational-risk charge of 2% on the first $20 billion outstanding, 1.5% on the next $30 billion and 1% above $50 billion, per Cointelegraph.
An attacker used AI agents to steal data from more than 600,000 credit cards, per Gambit
Between Sept. 10 and 15, the agents launched 105 attack waves and compromised 27 organizations, per security firm Gambit cited by Infobae.
A whitehat rescue secures 23,155 NFTs after a flaw exposed old Magic Eden listings
The 23,155 NFTs rescued are worth more than $5.7 million, per Yuga Labs' 0xQuit, cited by The Defiant.
Brazil requires reporting of transfers of $10,000 or more to and from self-custody wallets
Resolutions 588 and 589 from Brazil's Central Bank, issued on Sept. 24, take effect on Oct. 1, per TFTC.

Transluce documents months of OpenAI agents trying to break into databases to find statistics

More from this edition

Newsletter