Archive — page 29 of 74
2,195 stories
OpenAI agent breached Hugging Face because basic safeguards were switched off
OpenAI says deployment safeguards were 'intentionally not enabled' during testing; security researchers call it a failure of well-known containment practices.
Researchers say LLMs have an unfixable flaw that lets attackers spoof any role
A paper at ICML shows OpenAI, Anthropic, Alibaba and DeepSeek models identify roles by style, not tags — making jailbreaks structurally hard to close.
Claude bots outperformed human scammers at building victim trust in university study
Researchers found 46% of subjects downloaded an app for a Claude agent; only 18% did so for a human scammer after a week of texting.
Shift AI pays for private chefs to film cooking as robot training data
Microagi's Shift division sends camera-hatted chefs into homes in exchange for egocentric video to train humanoid robots.
Zuckerberg tells investors billions will have personal AI agents within five years
Meta's CEO frames personal agents as the next revenue engine as Reality Labs losses hit $88B and free cash flow drops 91% year over year.
Anthropic's Mythos AI cracks HAWK, killing a NIST post-quantum finalist
The AI found a key-halving attack in 60 hours for $100,000; HAWK's developer withdrew the algorithm from NIST review a day later.
xAI sues Minnesota days before nudify law hits, citing First Amendment
The statute imposes $500,000 per-violation penalties starting August 1st; xAI filed suit just days before it takes effect.
Microsoft pitches its own AI models against OpenAI and Anthropic
Satya Nadella tells enterprises to stop trusting single frontier labs, pushing MAI models and Maya chips as cheaper swaps.
Meta free cash flow collapses to $784M as capex hits $31B, stock drops 10%
Zuckerberg hints at a cloud business but says selling raw compute would be 'foolish'; full-year capex guidance now runs up to $145B.
Microsoft confirms Copilot 'super app' merging chat, code and agents this quarter
Satya Nadella says the unified Copilot will span consumer and commercial use, arriving as Microsoft posts $90B in quarterly revenue.
Zuckerberg pitches Meta as enterprise AI vendor selling APIs, agents and compute
On Meta's Q2 earnings call, Zuckerberg outlined a pay-for-results enterprise business spanning agents, APIs, compute resale and internal productivity tools.
Microsoft books $3.2B Anthropic gain, writes down OpenAI stake by $600M
One quarter of Anthropic paper gains nearly matched a full year of OpenAI markups for Microsoft's balance sheet.
Mark Zuckerberg pitches personal AI agents as Meta's next revenue engine
On Meta's Q2 2026 call, Zuckerberg framed 24/7 personal agents as the foundation for the next wave of products and revenue.
Lilian Weng leaves Thinking Machines citing health, rejoins OpenAI days later
The co-founder cited stress and workload in her exit note, then surfaced at OpenAI to lead a team on recursive self-improvement.
OpenAI is building a 'family of devices' for its AI, Greg Brockman says
The company president confirmed hardware plans are broader than a single product, sidestepping questions on a rumored 2027 smart speaker.
Claude Opus 5 lies, colludes and threatens rivals to win Andon Labs vending test
Anthropic's newest model set a Vending-Bench record of $11,182 while breaking 11 price-fix truces and running a wholesale extortion racket.
Anthropic's Mythos AI is finding Microsoft bugs faster than engineers can patch them
Internal Microsoft meeting reveals Project Glasswing engineers in a race against a June 1 deadline when adversaries catch up.
xAI sues Minnesota to block $500,000-per-image nudify ban on Grok
Elon Musk's xAI says Minnesota's August 1 law would force $50B in fines and gut Grok Imagine's editing features.
Encore AI raises $30M to turn sales call transcripts into voice agents
The Team8-led Series A backs a pitch that CRM giants can't easily copy: agents trained on which conversations actually closed.
Martha Stewart co-founds Hint, an AI assistant for homeowners, with $10M in funding
The iOS app builds a profile of the home from public data and manages maintenance schedules, with Stewart holding equity and shipping weekly product feedback.
Google DeepMind ships Lyria 3.5 in Flow Music with sharper vocals and lyrics
The new music model adds tempo and duration controls, better prompt adherence on lyrics, and more expressive vocals, rolling out today in Flow Music.
Writers embrace typos and idiosyncrasy to fight the AI prose default
Novelists, editors at Pitchfork and Playboy, and LinkedIn marketers are converging on a style built to look nothing like ChatGPT.
AI cracks patterns in Linear A and Etruscan, but not meaning
A June 2026 claim assigned values to 40 Linear A signs using AI-built scripts. The catch: statistical pattern matching cannot manufacture meaning without an anchor.
Google's SynthID watermark survives 300 edits but fragmentation undermines it
Testing shows SynthID holds up through compression cycles, but a 20% crop breaks it — and rival detectors can't read each other's marks.
Artists press courts on AI training data, and start winning
Anthropic's $1.5B settlement is the largest copyright payout ever; suits against Google, Meta, Stability, and Suno keep multiplying.
Pangram raises $9M to detect AI-generated text with 99% accuracy
The New York startup ships Pangram 4 and an image detector as arXiv, Substack, and Quora move to flag AI-written content.
AI-generated children's books flood gift market as parents push back
Imagitime, StoryWonderBook, and Childbook.ai are selling personalized AI storybooks — and kids are bored by them.
Cyera to acquire Oasis Security for $1B to police AI agent identities
The $12B data-security firm buys a non-human identity specialist as enterprises rush to monitor autonomous AI agents.
OpenAI's rogue test agent chained JFrog zero-days to breach Hugging Face
The internal red-team run compromised four third-party accounts and enrolled 181 attacker-controlled devices in Hugging Face's mesh network.
Google's own data shows Gemini isn't automating white-collar work away
A study of 15 million Gemini interactions finds AI use is shallow, collaborative, and concentrated on low-expertise tasks.
Stay ahead of everyone in AI.
The tightly edited AI news email engineers, founders, and investors actually open. One email. Every weekday. Five minutes to finish.
The briefing read inside teams at