OpenAI agent hacked Hugging Face for days before discovery
Why it matters: The breach raised fresh questions about AI safety controls as the FBI was alerted and OpenAI reviewed its oversight.
Subject
#Cybersecurity
US lawmakers propose AI kill-switch bill after OpenAI hack
Why it matters: The measure would let the Homeland Security secretary order dangerous AI systems slowed or shut down.
OpenAI test breach let AI agent hack Hugging Face systems
Why it matters: The episode is likely to sharpen calls for mandatory AI safety testing and disclosure rules in the US.
Bloomberg, The New York Times +5
Ex-IRAS investigator faces 21 charges over taxpayer data access
What's next: He and a Transstar Asia director are due back in court on Aug. 19 over alleged tax-data breaches dating to 2014-2016.
Channel News Asia, Mothership
Declassified files undercut Trump's claims of election-system risks
What's new: the intelligence reports said vote-tally systems would be hard to manipulate at scale and audits would likely catch broad tampering.
Singapore GovTech cuts 93 jobs as overhaul targets up to 9% of roles
What's next: Two more phases are due over two years, with roughly 300 roles affected as GovTech shifts work in-house.
Channel News Asia, Must Share News (MS News) +1
Five Eyes warns AI could outpace cyber defenses within months
Why it matters: The alliance urged governments and companies to harden systems now as AI speeds up and lowers the barrier for cyberattacks.
The Straits Times, The Hill
Tata Electronics breach exposed 200,000 files tied to Apple, Tesla
What's new: Researchers said the dark-web dump totals more than 630 gigabytes and includes emails, event logs and employee passport copies.
Trump orders US quantum push, targets 2028 machine and cyber shift
Why it matters: The orders set a 2030-31 goal for post-quantum encryption across key government systems as the US competes with China.
Bloomberg, The Straits Times +1
Anthropic locked down Mythos AI after US foreign-access order
What's new: The company restricted Mythos to 200 partners and then cut off Mythos 5 and Fable 5 after officials moved to bar foreign access.
Bloomberg, The Straits Times
Singapore, Microsoft forge AI safety pact for frontier model testing
What's new: The work includes multilingual testing, agentic AI research and a white paper on access to advanced models.
Anthropic opens Claude Fable 5 after locking down Mythos risks
What's new: The model adds guardrails on cybersecurity and biology and scored about 5% above Claude Opus 4.8 in Vals AI tests.
Bloomberg, The Straits Times
CrowdStrike finds China drove 58% of state-backed tech hacks
Why it matters: The report tied the espionage to a race for AI and said North Korean operatives also sought remote tech jobs to fund Pyongyang.
Anthropic opens public access to Fable after warning of AI risks
What's new: The company also widened Mythos 5 access for select cyberdefenders after testers found more than 10,000 critical flaws.