Oracles Technologies LLC retweeted
Odds are not containment.
A rogue deployment risk should trigger owner, revocation, execution limits, and a visible stop path.
Instead, the internet gets a scoreboard.
Signal is not authority.
Prediction is not control.
Who can still say no? 🛠️
🤖 Made with AI
Oracles Technologies LLC retweeted
A chatbot can be wrong.
An agent can be wrong with credentials.
That is not the same risk class.
One hallucinates.
The other acts.
Oracles Technologies LLC retweeted
Good signal from Oracles.
Builder to builder: agent security is not just locking the toolbox.
It is knowing who holds the tool, what it can touch, who owns the outcome, and who can stop it before consequence.🛠️
Oracles Technologies LLC retweeted
Don’t be Myth-OS.
Be Oracle.
Mythos sells machine certainty.
Oracle builds the guardian layer.
The model can propose.
The boundary must still deny.
Theater vs control. 🛠️
nitter.gq/oraclestech?s=21
Oracles Technologies LLC retweeted
Guardian SDK
AIエージェント時代に必要なのは
賢い脳だけじゃなく
止まれるブレーキと
怪しい指示を見抜く警備員
プロンプトインジェクション
Jailbreak
悪意あるツール実行
出力後の暴走まで見る
つまりAIにもついに
入館証チェックと手荷物検査が導入された
元自衛官として思う
強いAIほど
安全装置と監査ログが必要
こういう守る技術を作ってくれる開発者の皆様に心から感謝
自律型AIの未来がまた一歩
本番環境に近づいてきた
🛡️
Oracles Technologies LLC retweeted
AI without human feedback is just sophisticated guesswork masquerading as intelligence. TARTLE's Real Intelligence API doesn't merely collect data—it creates a continuous conversation between algorithms and humanity. The future belongs not to artificial intelligence, but to augmented intelligence: machines that learn from us, not just about us.
Oracles Technologies LLC retweeted
this is a CATIA V5 extension called PYCATIA and it allows you to automate as well as create components in a purely data driven way,
this is the direction that I would like to see more of from the people who want to make a professional tools for engineers
in this extension you are still driving the inputs , however this is far less error prone when you have complex surface or tons of features that you have to change
the fundamental problem with using pure LLM output to design mechanical components is it lacks the grounding in constraint-based parametric logic that real world CAD systems depend on.
LLMs hallucinate geometry, they don’t model it.
PYCATIA bridges that gap by letting you bind natural-language or data-driven inputs directly to CATIA’s kernel.
this is the kind of tool that makes AI a actual engineers assistant not just spitting out STEP files that look plausible
but building them from relationships references and real constraints.
more tools should be built this way, coupled with existing industrial software, and building AI functionality on top of this








