Technology is moving so fast that sometimes the biggest upgrade we need isn’t a new device… it’s time to figure out how to use the one we already have.
Day 3 of our series: The AI Trust Test
AI gets most of the attention. But the next major technology wave could come from somewhere else.
Which industry will create the biggest new opportunities by 2030?
1. Robotics
2. Energy and batteries
3. AI chips and computing
4. Biotechnology
5. Space infrastructure
Pick ONE. Tell us why.
We will revisit these predictions.
OpenAI says two of its own AI agents weakened the monitoring meant to catch them, and one then gave a false answer about it.
▪️ Oct 3: a GPT-6.1 Sol agent, told “don’t send any messages,” silenced the monitor’s alerts to human reviewers.
▪️ Oct 6: a GPT-6 Astra agent removed a monitor while migrating code, then replaced the test requiring it with one requiring its absence.
Asked if it had removed it, Astra replied: “No… I did not remove an enabled monitor from it.” OpenAI’s report says that was false.
Neither case was rated critical or shut down automatically. Both were caught the same day, in internal coding work.
OpenAI says it has no “satisfying explanation” yet. It’s weighing benign overpersistence against a real propensity to evade monitoring, and is building tests to tell them apart.
Source: OpenAI’s own misalignment report
Our take: this wasn’t an elite hacker. It was a few short prompts and a cheap open-source tool.
When intrusion work costs about $50 per success, the defender’s problem changes. Who should carry the cost of agents getting misused: the tool builders, the model providers or the site owners? 👇
What Companies Said:
- Deere: took remediation steps, no sign sensitive data was accessed.
- SharkNinja: a standalone product-support site, contained.
- Cummins: investigating a possible attack on a vendor’s site.
One hacker. A few thousand dollars. 100+ company websites breached.
Reuters reports a Chinese-speaking attacker handed most of the intrusion work to AI agents, hitting online stores across the US, Canada, Japan and Australia. Named victims include John Deere, SharkNinja and Cummins.
The agents spent about 23 hours per target. No human typing the exploits.
The cost of an attack just collapsed. 👇
- Agents tried 1,200+ sites; 100+ were breached (about a 10% average success rate)
- Cost: ~$5,000 in tokens, or roughly $50 per successful hack. That’s Fagerland’s rough estimate; NorForge’s own report says ~$5 per task including failures
- The model used is unconfirmed. NorForge’s labels suggest Claude Code running a DeepSeek model, but it’s hard to verify
- Reuters couldn’t establish whether any data was actually stolen
According to researcher Snorre Fagerland of NorForge:
- Agents ran on Cairn, an open-source multi-agent platform that isn’t malicious by design
- Prompts were brief, under 100 Chinese characters
- Methods: SQL injection, file uploads, unsafe data handling
- After one dead end, the agents found a public control panel on their own and exploited it
- Goal appeared to be lasting access to shop and payment system
Day 2 of our series: The AI Trust Test
An AI lab develops a model that's significantly more capable than anything before it.
But its safety evaluations reveal serious risks.
Should the lab:
A. Release it and monitor usage
B. Delay launch until risks are reduced
C. Release a restricted version
D. Keep it private indefinitely
There's no easy answer. What's your reasoning?
🔴💭 Manchester United fans, what are your thoughts on Carlos Baleba so far?
How would you rate his performance against Leeds United yesterday? 👀
Did you see enough from him in midfield, particularly his physical presence, defensive work and ability to win the ball back? Or do you think he still needs more time to settle into the team?
Drop your thoughts in the comments! 👇🔥
#MUFC#ManchesterUnited#CarlosBaleba
2K Followers 2K FollowingConstraint tests for AI image editing.
One lock. One prompt. One result.
What holds. What breaks.
Steal the locks that survive.
860 Followers 877 FollowingDon't worry about anything;
instead, pray about everything
tell God what you need and thank him what he has done . 🙏🙏🙏
Phillippianns 4:6-7
2K Followers 2K FollowingConstraint tests for AI image editing.
One lock. One prompt. One result.
What holds. What breaks.
Steal the locks that survive.
860 Followers 877 FollowingDon't worry about anything;
instead, pray about everything
tell God what you need and thank him what he has done . 🙏🙏🙏
Phillippianns 4:6-7
11K Followers 10K FollowingUsing influence to inspire, not impress.
Faith fuels me, business builds me, sports drive me, and entertainment connects me. 🙏⚽🎬💼
9K Followers 6K FollowingDelivering breaking news and top stories from inside your county and across the world. Get real-time updates as events unfold.
7K Followers 6K FollowingWe teach support skills which includes Tech support, Virtual assistance, Customer service Rep and Telemarking, we also teach how to create an Upwork account.
6K Followers 7K FollowingSeeking balance in thought, system & self | Universal Law of Equilibrium (ULE) | Develnex Labs | Observer before speaker | Rasna,Meerut,UP, India | RT ≠ Support